Open source website builder and Webflow alternative. Webstudio is an advanced visual builder that connects to any headless CMS, supports all CSS properties, and can be hosted anywhere, including with us.
TypeScript accessibilityalternativecloudflarecloudflare-workersdesignfreefrontendopen-sourceradix-uireactremixweb-developmentwebflowwebsite-builderwebsite-developmentwebsiteswebstudio
fix: Use writable npm cache for previews (#6272)
Oleg Isonen9ec4c820e180
diff --git a/docs/university/mcp.md b/docs/university/mcp.mdindex 3220891d18..9167d50629 100644--- a/docs/university/mcp.md+++ b/docs/university/mcp.md@@ -455,27 +455,29 @@ Vision-capable AI can use MCP to see what it is building:5. `preview.start` and `webstudio preview` install generated app dependencies under `.webstudio/preview` and reuse them across regenerations.6. Session previews download missing project assets into `.webstudio/assets`. If `PREVIEW_ASSET_DOWNLOAD_FAILED` occurs, restore network and project asset access, then retry `preview.start`.7. When launcher metadata is available, dependency installation reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.-8. When npm is selected, dependency installation honors `npm_config_cache`, including a caller-provided writable cache on Windows.-9. Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.-10. If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.-11. After MCP mutations, path-based screenshots regenerate the current session in place, wait for its exact project version, and normally reload the route. The server and browser remain alive. From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running generated site without starting it. Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview; a separate one-shot `single-op-call` process does not own another process's preview controller.-12. For multi-page work, capture each changed page by path through the same preview server, for example screenshot({ path: "/" }), screenshot({ path: "/pricing" }), and screenshot({ path: "/about" }). The screenshot tool navigates directly to the requested route; no browser click navigation is required.-13. For responsive work, call list-breakpoints first, then capture screenshots at viewport widths based on the Builder breakpoints plus a narrow mobile and desktop width.-14. Call screenshot with { path: "/" } or the changed page path and viewport such as { width: 375, height: 812 } and { width: 1440, height: 900 }. For an existing preview in another process, call screenshot with { baseUrl: "http://127.0.0.1:5177", path: "/" }. Use waitForSelector when the page has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.-15. An explicit occupied `port` fails immediately with `PREVIEW_PORT_IN_USE`. To capture a generated site already running in another process, pass its `baseUrl` with `path`; otherwise choose another port.-16. Automatic browser discovery checks system installations, configured browser paths, and Chromium installations in the Playwright browser cache.-17. The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.-18. When a baseline PNG exists, call screenshot.diff with baselinePath, currentPath, and outputDir for each page/viewport pair. Add expectedText when a specific visible phrase must be present; its assertions report pass/fail plus found and missing text. Add expectedVisual to set pass/fail limits for mismatch percentage, the number of changed regions, or an overall dominant color/brightness direction.-19. Read screenshot.diff textAnalysis: it reports OCR status plus text that appeared, disappeared, moved, changed content, or changed font/style geometry. If OCR is unavailable, expectedText assertions fail and textAnalysis reports why; ask the user for permission to install Tesseract, then call vision.install-ocr with { "confirm": true }, or rely on visual inspection.-20. Inspect every viewport PNG and any diff artifacts with vision, then compare layout, OCR text evidence, color, spacing, imagery, and responsive framing against the user intent.-21. If the screenshot does not match, apply another focused mutation and repeat screenshot verification.+8. When npm is selected, dependency installation honors `npm_config_cache`. If it is unset, preview uses the writable `.webstudio/preview/.npm-cache` directory instead of the user's default npm cache.+9. For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.+10. Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.+11. If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.+12. After MCP mutations, path-based screenshots regenerate the current session in place, wait for its exact project version, and normally reload the route. The server and browser remain alive. From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running generated site without starting it. Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview; a separate one-shot `single-op-call` process does not own another process's preview controller.+13. For multi-page work, capture each changed page by path through the same preview server, for example screenshot({ path: "/" }), screenshot({ path: "/pricing" }), and screenshot({ path: "/about" }). The screenshot tool navigates directly to the requested route; no browser click navigation is required.+14. For responsive work, call list-breakpoints first, then capture screenshots at viewport widths based on the Builder breakpoints plus a narrow mobile and desktop width.+15. Call screenshot with { path: "/" } or the changed page path and viewport such as { width: 375, height: 812 } and { width: 1440, height: 900 }. For an existing preview in another process, call screenshot with { baseUrl: "http://127.0.0.1:5177", path: "/" }. Use waitForSelector when the page has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.+16. An explicit occupied `port` fails immediately with `PREVIEW_PORT_IN_USE`. To capture a generated site already running in another process, pass its `baseUrl` with `path`; otherwise choose another port.+17. Automatic browser discovery checks system installations, configured browser paths, and Chromium installations in the Playwright browser cache.+18. The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.+19. When a baseline PNG exists, call screenshot.diff with baselinePath, currentPath, and outputDir for each page/viewport pair. Add expectedText when a specific visible phrase must be present; its assertions report pass/fail plus found and missing text. Add expectedVisual to set pass/fail limits for mismatch percentage, the number of changed regions, or an overall dominant color/brightness direction.+20. Read screenshot.diff textAnalysis: it reports OCR status plus text that appeared, disappeared, moved, changed content, or changed font/style geometry. If OCR is unavailable, expectedText assertions fail and textAnalysis reports why; ask the user for permission to install Tesseract, then call vision.install-ocr with { "confirm": true }, or rely on visual inspection.+21. Inspect every viewport PNG and any diff artifacts with vision, then compare layout, OCR text evidence, color, spacing, imagery, and responsive framing against the user intent.+22. If the screenshot does not match, apply another focused mutation and repeat screenshot verification.Generated app setup:- `preview.start` and `webstudio preview` install generated app dependencies under `.webstudio/preview` and reuse them across regenerations.- Session previews download missing project assets into `.webstudio/assets`. If `PREVIEW_ASSET_DOWNLOAD_FAILED` occurs, restore network and project asset access, then retry `preview.start`.- When launcher metadata is available, dependency installation reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.-- When npm is selected, dependency installation honors `npm_config_cache`, including a caller-provided writable cache on Windows.+- When npm is selected, dependency installation honors `npm_config_cache`. If it is unset, preview uses the writable `.webstudio/preview/.npm-cache` directory instead of the user's default npm cache.+- For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.- If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.diff --git a/packages/cli/src/commands/preview.test.ts b/packages/cli/src/commands/preview.test.tsindex 5a709fa51f..67fb4e9d93 100644--- a/packages/cli/src/commands/preview.test.ts+++ b/packages/cli/src/commands/preview.test.ts@@ -495,6 +495,7 @@ test("installs isolated generated dependencies when the cli does not ship them",cwd: "/tmp/project/.webstudio/preview",env: expect.objectContaining({PATH: "/opt/webstudio-node/bin:/usr/bin",+ npm_config_cache: "/tmp/project/.webstudio/preview/.npm-cache",}),timeout: 120_000,})diff --git a/packages/cli/src/commands/preview.ts b/packages/cli/src/commands/preview.tsindex 1f770e80d1..f720daeae5 100644--- a/packages/cli/src/commands/preview.ts+++ b/packages/cli/src/commands/preview.ts@@ -107,6 +107,8 @@ export const previewOptions = (yargs: CommonYargsArgv) =>"Preview dependencies are isolated under .webstudio/preview and reused across regenerations.","Do not add generated-preview dependencies to the repository root package.json or pnpm-lock.yaml.","When launcher metadata is available, Preview reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.",+ "Unless npm_config_cache is already configured, npm uses a writable cache inside .webstudio/preview.",+ "For npm cache permission errors, unset npm_config_cache to use the preview-local cache, then retry.","If dependency installation fails, check the reported package-manager path and network configuration, then reinstall or update the Webstudio CLI if the problem persists.",].join("\n"));@@ -380,20 +382,31 @@ export const ensurePreviewDependencies = async (invocation.command === operations.nodeExecPath? invocation.args[0]: invocation.command;+ const runtimeEnv = getNodeRuntimeEnv(+ operations.nodeExecPath,+ operations.env,+ operations.platform+ );+ const installEnv =+ packageManager.name === "npm" &&+ Object.keys(runtimeEnv).every(+ (name) => name.toLowerCase() !== "npm_config_cache"+ )+ ? {+ ...runtimeEnv,+ npm_config_cache: join(previewProjectDir, ".npm-cache"),+ }+ : runtimeEnv;try {await operations.execFile(invocation.command, invocation.args, {cwd: previewProjectDir,- env: getNodeRuntimeEnv(- operations.nodeExecPath,- operations.env,- operations.platform- ),+ env: installEnv,timeout: previewDependencyInstallTimeout,});} catch (error) {throw new Error(- `PREVIEW_DEPENDENCY_INSTALL_FAILED: Could not install the generated preview dependencies with ${packageManager.name} at ${packageManagerPath}. Check the package-manager/network configuration, then reinstall or update webstudio if the problem persists.\n\nPackage-manager diagnostics:\n${getPreviewInstallFailureDiagnostics(error)}`,+ `PREVIEW_DEPENDENCY_INSTALL_FAILED: Could not install the generated preview dependencies with ${packageManager.name} at ${packageManagerPath}. For npm cache permission errors, unset npm_config_cache to use the preview-local cache, then retry. Check the package-manager/network configuration, then reinstall or update webstudio if the problem persists.\n\nPackage-manager diagnostics:\n${getPreviewInstallFailureDiagnostics(error)}`,{ cause: error });}diff --git a/packages/cli/src/docs.generated.ts b/packages/cli/src/docs.generated.tsindex 3a8dd1649d..2faf5aadb7 100644--- a/packages/cli/src/docs.generated.ts+++ b/packages/cli/src/docs.generated.ts@@ -2,7 +2,7 @@// Edit markdown files in src/docs instead.export const cliDocs = {"api-use-cases":- '# CLI API Use Cases\n\n## Link/configure one project\n\nCommands:\n\n- webstudio init --link <api-share-link> --json\n\nNotes:\n\n- Writes local project id and global origin/token config.\n\n## Import synced project bundle into another project\n\nCommands:\n\n- webstudio sync\n- webstudio import --to <destination-share-link>\n- MCP tool: import {"to":"<destination-share-link>"}\n\nNotes:\n\n- Imports local `.webstudio/data.json` into the destination project.\n- Destination share link must allow build/import access.\n- Use `--skip-assets` only when asset rows and files should not be imported.\n\n## Identify current token\n\nCommands:\n\n- MCP tool: whoami {}\n\n## Check token permissions\n\nCommands:\n\n- webstudio permissions --json\n\n## Inspect project/build/version\n\nCommands:\n\n- MCP tool: inspect {}\n- MCP tool: snapshot {"include":["pages","instances","styles"]}\n\n## Discover CLI/API capabilities\n\nCommands:\n\n- webstudio schema api\n- webstudio schema mcp\n- webstudio man --json\n- webstudio man llm --json\n- MCP tool: meta.index {}\n- MCP tool: meta.guide {"brief":"Create a pricing page"}\n- MCP tool: meta.get-more-tools {"brief":"update-styles"}\n- webstudio mcp list-resources\n- webstudio mcp read-resource webstudio://project/guide\n- webstudio mcp read-resource webstudio://project/expressions\n\nNotes:\n\n- Use `webstudio schema mcp` for a compact machine-readable MCP tool overview. Add `--verbose` or use focused `meta.get-more-tools` calls only when exact input schemas are needed.\n- Use focused MCP tools for discovery first: `meta.index`, `meta.guide`, `meta.get-more-tools`, `components.list`, `components.summary`, `components.search`, `components.get`, `templates.list`, and `templates.get`. Protocol clients can use `resources/list` and `resources/read`; shell agents can use `webstudio mcp list-resources` and `webstudio mcp read-resource <uri>`. Read longer resources such as `webstudio://project/tools` and `webstudio://project/components` only when focused tools are insufficient.\n- `components.summary` returns counts by default; request `{"detail":"components","limit":20}` for paginated entries. Registry list tools return compact metadata, while `components.get` and `templates.get` return focused full details.\n- Read `webstudio://project/expressions` before authoring unfamiliar computed text, prop bindings, resource expressions, actions, or Collection item bindings.\n- From a shell, call one MCP tool with the shortcut form `webstudio <tool> \'<json>\'`, for example `webstudio components.summary`. The explicit equivalent is `webstudio mcp single-op-call <tool> \'<json>\'`. Use `--input-file` for large payloads.\n\n## Inspect external shadcn registry items\n\nCommands:\n\n- webstudio registry inspect --source https://example.com/r/registry.json --item button --json\n- webstudio registry inspect --source ./registry.json --item dialog --json\n- webstudio registry inspect --source https://example.com/r/button.json --json\n\nNotes:\n\n- Reads a local or remote registry item without installing files or changing the configured Webstudio project.\n- Returns the item name, description, package and registry dependencies, file paths/targets, available docs, and a read-only compatibility report.\n- The report explicitly says whether installation or editable-component conversion is supported, lists declared requirements and manual steps, and says when arbitrary source code was not analyzed.\n- This is an inspection step only. It does not install files or change the configured project.\n\n## Understand what MCP can do\n\nCommands:\n\n- webstudio man mcp\n- MCP tool: meta.index {}\n- MCP tool: meta.guide {"brief":"What can Webstudio MCP do?"}\n\nMCP lets agents work on one configured Webstudio project. Agents can:\n\n- Inspect the linked project, token permissions, and latest editable build.\n- Read selected project data for audits, migrations, and repair.\n- Search values and ids across every Builder namespace without sending full project data to the model.\n- Audit accessibility, security, SEO, performance settings, unused assets, ineffective Collection styles, and unused or duplicate style data.\n- Create and edit pages, folders, redirects, breakpoints, and page templates.\n- Create pages from reusable templates.\n- Update page metadata, SEO fields, auth settings, and marketplace metadata.\n- Insert components and styled JSX sections.\n- Create data-driven lists, grids, cards, and similar repeated UI from array or object data in one Collection operation.\n- Move, copy, wrap, unwrap, convert, rename, retag, and delete elements.\n- Update text, rich text, props, bindings, and actions.\n- Create and update local styles, design tokens, style sources, and CSS variables.\n- Create static data variables and JSON variables.\n- Create HTTP, GraphQL, and system resources.\n- Use system resources for sitemap, current date, and assets.\n- Bind resources to rendered data or form/action props.\n- Manage nested asset folders and upload, inspect, move, duplicate, download, replace, delete, and inspect usage for assets.\n- Publish, unpublish, inspect publish jobs, and manage custom domains.\n- Start preview, capture screenshots, compare screenshot diffs, and use OCR when installed.\n\n## Inspect and refresh MCP session cache\n\nCommands:\n\n- MCP tool: status {}\n- MCP tool: status {"verbose":true}\n- MCP tool: refresh {"namespaces":["pages","instances","styles"]}\n- MCP tool: reset-session {}\n\nNotes:\n\n- Use status before a task to understand the cached ProjectSession state.\n- Use status with `{"verbose":true}` only when debugging full namespaces, freshness, compatibility, or diagnostics.\n- Use refresh when project data may have changed outside the current MCP session.\n- Use reset-session when local cached state is corrupt or incompatible.\n\n## Visually verify rendered work with AI vision\n\nCommands:\n\n- MCP tool: preview.start {"host":"127.0.0.1","port":5173}\n- MCP tool: preview.status {}\n- MCP tool: preview.stop {}\n- MCP tool: screenshot {"path":"/","output":".webstudio/screenshots/home-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"path":"/pricing","output":".webstudio/screenshots/pricing-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"baseUrl":"http://127.0.0.1:5177","path":"/pricing","output":".webstudio/screenshots/pricing-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff"}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff","expectedText":["Pricing","Start free"]}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff","expectedVisual":{"maxMismatchPercentage":2,"maxChangedRegions":3,"dominantColorChange":{"channel":"luminance","direction":"increase","minMagnitude":10}}}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/pricing-before.png","currentPath":".webstudio/screenshots/pricing-current.png","outputDir":".webstudio/screenshots/diff"}\n- MCP tool: vision.install-ocr {"confirm":true}\n\nNotes:\n\n- Enter this workflow only when the user explicitly requests visual verification or opts in after being asked. Do not start preview, screenshots, diffs, OCR, or rendered audits automatically after a mutation.\n- `preview.status` reports whether generated output is `stale`. When no preview is running, `url`, `pid`, and `mode` are omitted. When present, `renderedProjectVersion` is the last project version materialized into the preview.\n- A managed `screenshot` or another `preview.start` refreshes stale generated output before capture.\n\n- After opt-in, use this so a vision-capable AI can see the generated site from the current MCP session. Use `path`; never pass a Webstudio Builder/share URL or capture Builder chrome.\n- For multi-page work, capture every changed page by `path` through the same preview server; no click navigation is required.\n- Iterative mode is the default: after MCP mutations, path screenshots ensure generated project files are current, wait for the exact session version, and perform an ordinary page reload without Vite HMR. The preview server and browser stay alive. Use `{"mode":"production"}` only for release-like verification; rendered audit does this automatically.\n- Calling `preview.start` after a committed mutation restarts a stale iterative server so external browsers and HTTP clients receive the newly generated project.\n- Do not call `preview.start` through one-shot `webstudio mcp single-op-call`: it is long-lived. From a shell, use `webstudio mcp run` with preview.start, screenshot, and preview.stop in one shared process, or use a real long-running MCP client.\n- From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running preview/site without generating, building, starting, or restarting preview.\n- Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview. A separate one-shot `single-op-call` process does not own another process\'s preview controller.\n- Use waitForSelector when the rendered app has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.\n- The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.\n- Preview installs generated app dependencies under `.webstudio/preview` and reuses them across regenerations.\n- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.\n- When launcher metadata is available, Preview reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.\n- If dependency installation fails, check the reported package-manager path and network configuration, then reinstall or update the Webstudio CLI if the problem persists.\n- When a baseline exists, use screenshot.diff once per baseline/current page or viewport pair to get changed regions, OCR textAnalysis, and diff artifact paths before deciding whether the result matches. Pass expectedText for explicit pass/fail current-screen text assertions with found and missing text. Pass expectedVisual for pass/fail limits on pixel mismatch percentage, changed-region count, or the overall dominant color/brightness direction.\n- If screenshot.diff reports OCR unavailable and the user agrees to install it, call vision.install-ocr {"confirm":true}; otherwise continue with pixel diff and visual inspection.\n- Compare the PNG, OCR text evidence, and diff artifacts against the user\'s intent for layout, typography, colors, spacing, imagery, and responsive framing; then iterate with focused mutations.\n- Root CLI equivalent: `webstudio screenshot --path /pricing --output pricing.png` generates a temporary production preview, captures that route, and stops the server. For repeated captures, keep `webstudio preview` running and pass its absolute URL to `webstudio screenshot`.\n\n## List pages\n\nCommands:\n\n- MCP tool: list-pages {}\n- MCP tool: list-folders {}\n\n## Read page by id\n\nCommands:\n\n- MCP tool: get-page {"pageId":"<pageId>"}\n\n## Read page by path\n\nCommands:\n\n- MCP tool: get-page-by-path {"path":"/pricing"}\n\n## Create page\n\nCommands:\n\n- MCP tool: create-page {"name":"Pricing","path":"/pricing"}\n- MCP tool: create-page {"name":"Pricing","path":"/pricing","title":"Pricing","meta":{"description":"Plans for teams"}}\n\nNotes:\n\n- `name`, `path`, page `title`, and metadata text fields accept plain fixed values.\n- For computed page titles or metadata, send JavaScript expression code such as `pageTitle ?? "Pricing"`.\n\n## Update page settings/metadata\n\nCommands:\n\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans","status":200}}}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"meta":{"auth":{"method":"basic","login":"<login>","password":"<password>"}}}}\n\nNotes:\n\n- Page `title` and metadata text fields accept plain fixed values.\n- For computed page titles or metadata, send JavaScript expression code such as `pageTitle ?? "Pricing"`.\n- Page `status` accepts a fixed HTTP status code as a number from 200 through 599 or a JavaScript expression string for a dynamic status.\n\n## Read project settings\n\nCommands:\n\n- MCP tool: get-project-settings {}\n\nNotes:\n\n- Read `meta.agentInstructions` before making project changes. It contains the project\'s own guidance for AI agents.\n- Agent instructions are shared project guidance. Do not store credentials or other secrets there.\n\n## Update project settings\n\nCommands:\n\n- MCP tool: update-project-settings {"meta":{"siteName":"Acme"}}\n- MCP tool: update-project-settings {"meta":{"agentInstructions":"Use existing design tokens and keep product copy concise."}}\n\n## Read marketplace product\n\nCommands:\n\n- MCP tool: get-marketplace-product {}\n\n## Update marketplace product\n\nCommands:\n\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"asset-id","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n\n## Submit marketplace product\n\nCommands:\n\n- MCP tool: upload-asset {"asset":{"name":"marketplace-thumbnail.png","type":"image","format":"png","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"<uploadedAssetId>","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n- MCP tool: publish {"target":"production"}\n- MCP tool: submit-marketplace-product {"acknowledgePublicSubmission":true}\n\nNotes:\n\n- Wait for the production publish to complete before submitting the product for review.\n- Submission requires complete, valid marketplace metadata.\n\n## List redirects\n\nCommands:\n\n- MCP tool: list-redirects {}\n\n## Create redirect\n\nCommands:\n\n- MCP tool: create-redirect {"old":"/old","new":"/new","status":301}\n\n## Update redirect\n\nCommands:\n\n- MCP tool: update-redirect {"old":"/old","values":{"new":"/newer","status":302}}\n- MCP tool: update-redirect {"old":"/old","values":{"status":null}}\n\n## Delete redirect\n\nCommands:\n\n- MCP tool: delete-redirect {"old":"/old"}\n\n## Set redirects\n\nCommands:\n\n- MCP tool: set-redirects {"redirects":[{"old":"/old","new":"/new","status":"301"}]}\n\n## List breakpoints\n\nCommands:\n\n- MCP tool: list-breakpoints {}\n\n## Create breakpoint\n\nCommands:\n\n- MCP tool: create-breakpoint {"label":"Tablet","maxWidth":991}\n\n## Update breakpoint\n\nCommands:\n\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"label":"Tablet","maxWidth":1023}}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"condition":null,"minWidth":768}}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"minWidth":null,"maxWidth":null,"condition":"(hover: hover)"}}\n\n## Delete breakpoint\n\nCommands:\n\n- MCP tool: delete-breakpoint {"breakpointId":"tablet"}\n\n## Duplicate page\n\nCommands:\n\n- MCP tool: duplicate-page {"pageId":"<pageId>","name":"Pricing Copy","path":"/pricing-copy"}\n- MCP tool: duplicate-page {"pageId":"<pageId>","name":"Paris","path":"/paris","substitutions":{"text":{"London":"Paris"},"variables":{"city":{"type":"string","value":"Paris"}}}}\n- webstudio duplicate-page --page <pageId> --name Paris --path /paris --substitutions \'{"text":{"London":"Paris"},"variables":{"city":{"type":"string","value":"Paris"}}}\' --json\n\nNotes:\n\n- Text substitutions replace exact fixed text only in the duplicated page\'s text children, string props, title, and metadata.\n- Variable substitutions are keyed by copied source-variable name and use typed variable values. The operation rejects missing or ambiguous names without committing a partial duplicate.\n- Existing expressions and cloned variable/resource references keep their remapped ids.\n\n## List page templates\n\nCommands:\n\n- MCP tool: list-page-templates {}\n\n## Create page template\n\nCommands:\n\n- MCP tool: create-page-template {"name":"Landing Template","title":"Landing"}\n\n## Update page template\n\nCommands:\n\n- MCP tool: update-page-template {"templateId":"<templateId>","values":{"name":"Article Template","meta":{"description":"Reusable article layout"}}}\n\n## Delete page template\n\nCommands:\n\n- MCP tool: delete-page-template {"templateId":"<templateId>"}\n\n## Duplicate page template\n\nCommands:\n\n- MCP tool: duplicate-page-template {"templateId":"<templateId>"}\n\n## Reorder page template\n\nCommands:\n\n- MCP tool: reorder-page-template {"sourceTemplateId":"<sourceTemplateId>","targetTemplateId":"<targetTemplateId>","position":"before"}\n\n## Create page from template\n\nCommands:\n\n- MCP tool: create-page-from-template {"templateId":"<templateId>","name":"Landing","path":"/landing"}\n\n## Delete page\n\nCommands:\n\n- MCP tool: delete-page {"pageId":"<pageId>"}\n\n## List folders\n\nCommands:\n\n- MCP tool: list-folders {}\n- MCP tool: list-pages {}\n\n## Create folder\n\nCommands:\n\n- MCP tool: create-folder {"name":"Blog","slug":"blog"}\n\n## Update folder\n\nCommands:\n\n- MCP tool: update-folder {"folderId":"<folderId>","values":{"name":"Blog","slug":"blog"}}\n\n## Delete folder\n\nCommands:\n\n- MCP tool: delete-folder {"folderId":"<folderId>"}\n\n## List element instances\n\nCommands:\n\n- MCP tool: list-instances {"pagePath":"/","maxDepth":3}\n\n## Inspect one element instance\n\nCommands:\n\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["props","styles","children"]}\n\n## Insert authored JSX or one component template\n\nCommands:\n\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.element ws:tag=\'section\' ws:style={css`padding: 32px;`}><ws.element ws:tag=\'h2\'>Product OS</ws.element><radix.Switch><radix.SwitchThumb /></radix.Switch></ws.element>"}\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"@webstudio-is/sdk-components-react-radix:Switch"}\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"Form"}\n\nNotes:\n\n- Use MCP `insert-fragment` as the default way to author styled component trees. It converts JSX to a structured fragment before mutation.\n- Use only exact component ids returned by `components.search`, `components.get`, or `templates.get`. Never derive or guess component ids.\n- The `ws:` namespace contains specific Webstudio core components; it is not HTML-tag shorthand. Use `<ws.element ws:tag="div">` for a native `div` and `<ws.element ws:tag="form">` for a native form, never `<ws.div>` or `<ws.form>`.\n- For Webstudio\'s complete form structure, discover the Form component and insert its automatic template with `insert-component` using component `"Form"`.\n- MCP receives JSX as a JSON string because MCP arguments are JSON. The CLI converts it locally before the runtime mutation, so the project session receives structured Webstudio data, not JSX source.\n- In `insert-fragment` JSX, use ``ws:style={css`...`}`` for Webstudio-native CSS, or use React-style object syntax such as `style={{ padding: 24 }}` when that is simpler. Both forms create editable Webstudio style data.\n- Do not access host globals or dynamic code APIs in JSX fragments, including `process`, `globalThis`, `eval`, `Function`, or `constructor`.\n- Use Webstudio prop names such as `class` and `for`; do not use React aliases `className` or `htmlFor`.\n- Use Webstudio actions for event/action props, for example `onClick={new ActionValue(["event"], expression\\`console.log(event)\\`)}`. Do not pass JavaScript functions such as `onClick={() => ...}`.\n- Plain prop values must be JSON-compatible: `null`, strings, booleans, finite numbers, arrays, and plain objects. Do not pass `undefined`, `Symbol`, `BigInt`, `NaN`, `Infinity`, `Date`, `Map`, `Set`, class instances, or circular objects; omit the prop, use plain data, or use `expression`/`ActionValue` when the value is dynamic.\n- Template-backed components used in JSX must include required child/part components explicitly under the same parent structure as the template, for example `<radix.Switch><radix.SwitchThumb /></radix.Switch>`.\n- Webstudio applies a registered template automatically when using `insert-component`, so composed components such as Switch include required child parts and styles.\n- Use `components.list`, `components.summary`, `components.search`, `components.get`, `templates.list`, and `templates.get` to discover known registry items, component ids, props, templates, insertability, and content model. Read `webstudio://project/components` only when those focused tools are insufficient.\n- Component/template registry items use a shadcn-compatible top-level shape plus Webstudio-specific superset metadata in `meta`. They are for Builder/MCP discovery, not a published shadcn install registry yet.\n- Known components with `contentModel.category: "none"` are not standalone-insertable; insert their root component template instead so required providers/parents are included.\n- Unknown custom component ids are a low-level extension mechanism, not a discovery fallback. Agents must not synthesize them.\n\n## Make a region editable in Content mode\n\nCommands:\n\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"ws:block"}\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["children"]}\n\nNotes:\n\n- When a page will be handed to a Content-mode editor, wrap every region they should be able to edit in a Content Block (`ws:block`). Content-mode editors can edit text and supported props only in Content Block descendants. Content outside those blocks remains read-only, even when it looks like ordinary editable text.\n- Put reusable insertable options inside the Content Block\'s `ws:block-template` child. A template is source material, not editor content: editors cannot edit or delete it directly. When an editor inserts a template, its copy becomes a direct child of the Content Block and is editable.\n- Before handing off a page, verify with `inspect-instance` that the intended text, images, and links are inside a Content Block, and that templates include all required styling because Content-mode editors cannot use the Style panel.\n\n## Store Content Block content in MDX\n\nCommands:\n\n- MCP tool: list-assets {"type":"file"}\n- MCP tool: upload-asset {"asset":{"name":"article.mdx","type":"file","format":"mdx","meta":{}},"assetsDir":".webstudio/assets"}\n- MCP tool: connect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":{"type":"asset","assetId":"<mdxAssetId>"}}\n- MCP tool: switch-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":{"type":"asset","assetId":"<otherMdxAssetId>"}}\n- MCP tool: inspect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n- MCP tool: edit-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":"# Article title\\n\\nArticle body."}\n- MCP tool: update-content-block-frontmatter {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","properties":{"title":"Article title","draft":false}}\n- MCP tool: reload-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n- MCP tool: migrate-content-block-template-references {"assetIds":["<mdxAssetId>"],"migration":{"type":"rename","from":"Old template name","to":"New template name"}}\n- MCP tool: disconnect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n\nNotes:\n\n- Create the local `.mdx` file under `.webstudio/assets` before calling `upload-asset`. Use the returned Asset ID when connecting it.\n- Use the Content Block source operations for connecting, switching, inspecting, editing, reloading, and disconnecting. Do not create or delete the `src` prop with generic prop tools; the source operations preserve the Body outlet and Content Block lifecycle.\n- `renderScope` is a stable key for the rendered occurrence. Use a page-based key for a direct Content Block. For a repeated Collection occurrence, use a distinct key and pass the scoped values required to resolve an expression source. For example, use `source:{"type":"expression","value":"post.assetId"}` with `variables:{"post":{"assetId":"<mdxAssetId>"}}`.\n- Connecting replaces existing Body content. If the result returns `requiresConfirmation:true`, report that replacement to the user and repeat the same call with `confirmReplacement:true` only after approval.\n- Prefer Markdown whenever it can represent the component and all authored properties. Use `<ws.element ws:tag="tag">` for a standard HTML element with authored properties that Markdown cannot express. Use `<ws.element ws:name="Template name">` only for a uniquely named top-level Content Block template.\n- Read the Content Block\'s Templates children before writing `ws:name`. The value must exactly match a unique top-level template instance name. Preserve unresolved names and report their diagnostics instead of deleting them.\n- When a template is renamed or deleted, use `migrate-content-block-template-references` to update a selected set of affected MDX files. The first call returns a plan. Report its changed-file, update, omission, and diagnostic counts, then repeat the exact request with its `confirmationToken` only after approval. A rename changes matching `ws:name` values; a removal deletes matching MDX elements. Invalid files remain unchanged and are reported in diagnostics.\n- `edit-content-block-source` replaces the complete MDX source. Preserve frontmatter and unrelated source when making a bounded edit.\n- `update-content-block-frontmatter` replaces the complete frontmatter mapping. Inspect the current source first and include every property that must remain.\n- Use `update-content-block-frontmatter` for MCP frontmatter edits. MDX-rendered elements are not persistent instance targets for generic `bind-props` or `update-text` calls. Preserve existing `mode:"readwrite"` bindings when encountered; they are valid only for exact direct paths into the connected document\'s frontmatter. Keep computed expressions and `$ref` values read-only.\n- Inspect every returned diagnostic. Invalid MDX is saved rather than silently repaired; preserve the source, report the source range, and fix only the requested or invalid part.\n- If an edit in a long-lived MCP session reports a conflict after another client saved the Asset, call `reload-content-block-source`, inspect the latest source, reapply the requested change, and retry. One-shot CLI calls refresh before each operation and normally cannot reproduce a stale session. Never overwrite the newer revision blindly.\n\n## Move elements\n\nCommands:\n\n- MCP tool: move-instance {"moves":"moves.json contents"}\n\nNotes:\n\n- Use `position: "end"` to append an instance. Repeating this for A and then B preserves the final order A, B.\n- A numeric `insertIndex` addresses the target parent\'s children before the moved instance is removed. Use it for exact placement; do not calculate the last index to append.\n- Moves in one `moves` array are applied sequentially in array order.\n\n## Clone element subtree\n\nCommands:\n\n- MCP tool: clone-instance {"sourceInstanceId":"<instanceId>","targetParentInstanceId":"<targetParentId>"}\n\n## Delete element subtree\n\nCommands:\n\n- MCP tool: delete-instance {"instanceIds":["<instanceId>"]}\n\n## List text/expression children\n\nCommands:\n\n- MCP tool: list-texts {"pagePath":"/"}\n\n## Update text child\n\nCommands:\n\n- MCP tool: update-text {"instanceId":"<instanceId>","childIndex":0,"text":"Launch faster"}\n\n## Replace bounded literal text\n\nCommands:\n\n- MCP tool: replace-text {"find":"Start free","replace":"Get started","match":"exact","pagePath":"/pricing","limit":20}\n\nNotes:\n\n- This changes only literal text children, never expression children. Scope it to pagePath or pageId and set a limit before a broad replacement.\n\n## Replace bounded static prop text\n\nCommands:\n\n- MCP tool: replace-prop-text {"find":"old.example.com","replace":"www.example.com","match":"substring","names":["href","code"],"limit":20}\n\nNotes:\n\n- This changes only static string props such as href, alt, aria-label, title, and HTML embed code. It never changes expressions, resources, actions, assets, or other dynamic bindings. Use names or instanceIds and a limit to narrow the change.\n\n## Replace bounded resource text\n\nCommands:\n\n- MCP tool: replace-resource-text {"find":"api.old.example.com","replace":"api.example.com","fields":["url"],"limit":20}\n\nNotes:\n\n- This changes resource names and fixed URL literals only. It skips dynamic URL expressions, headers, search parameters, request bodies, and GraphQL query code.\n\n## Update props\n\nCommands:\n\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: replace-prop-text {"find":"Old label","replace":"New label","names":["aria-label","title"],"limit":20}\n\nNotes:\n\n- Use this for fixed prop values such as `aria-label`, `alt`, `id`, static `href`, and other direct string/number/boolean/json prop values.\n\n## Add JSON-LD structured data\n\nCommands:\n\n- MCP tool: components.get {"component":"JsonLd"}\n- MCP tool: insert-component {"parentInstanceId":"<headSlotInstanceId>","component":"JsonLd"}\n- MCP tool: update-props {"updates":[{"instanceId":"<jsonLdInstanceId>","name":"code","type":"string","value":"{\\"@context\\":\\"https://schema.org\\",\\"@type\\":\\"Organization\\",\\"name\\":\\"Acme\\"}"}]}\n- MCP tool: bind-props {"bindings":[{"instanceId":"<jsonLdInstanceId>","name":"code","binding":{"type":"expression","value":"({ \'@context\': \'https://schema.org\', \'@type\': \'Article\', headline: post.title })"}}]}\n- MCP tool: audit {"scopes":["seo"],"pagePath":"/"}\n\nNotes:\n\n- Prefer placing `JsonLd` inside `HeadSlot`.\n- For fixed structured data, store `code` as a JSON object or array encoded as a compact string. The Builder formats it for editing.\n- For structured data containing runtime values, use `bind-props` with an expression that evaluates directly to an object or array. Do not call `JSON.stringify` or assemble JSON with string concatenation. Webstudio stores the expression as source text, evaluates it at runtime, and the `JsonLd` component validates and serializes the resulting value.\n- The semantic prop update rejects malformed JSON and structurally invalid fixed JSON-LD with a precise JSON path.\n- The SEO audit also warns about a missing top-level `@context`, unknown or superseded Schema.org terms, properties unsupported by the supplied type, and incompatible primitive value types.\n- Schema.org vocabulary findings are warnings because custom vocabularies and extensions remain valid. Dynamic JSON-LD is marked as skipped for rendered validation.\n- Do not use bindings just to set static text.\n\n## Delete props\n\nCommands:\n\n- MCP tool: delete-props {"deletions":"props.json contents"}\n\n## Bind props to expressions/resources/actions\n\nCommands:\n\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n\nNotes:\n\n- Use this only when the prop should remain dynamic: expression, resource, action, or an existing scoped runtime context value such as `system`.\n- For a fixed string value, use `update-props` with `type:"string"` and a direct `value` instead.\n\n## Read styles\n\nCommands:\n\n- MCP tool: get-styles {"instanceIds":["<instanceId>"],"includeTokens":true}\n\n## Update local styles\n\nCommands:\n\n- MCP tool: update-styles {"updates":"styles.json contents"}\n\n## Delete local styles\n\nCommands:\n\n- MCP tool: delete-styles {"deletions":"styles.json contents"}\n\n## Replace matching style values\n\nCommands:\n\n- MCP tool: replace-styles {"property":"color","fromValue":{"type":"keyword","value":"red"},"toValue":{"type":"keyword","value":"blue"}}\n\n## List design tokens\n\nCommands:\n\n- MCP tool: list-design-tokens {}\n- MCP tool: list-design-tokens {"withUsage":true}\n- MCP tool: list-design-tokens {"verbose":true}\n\nNotes:\n\n- The default response is compact and includes token id, name, declaration count, and optional usage count.\n- Use `verbose:true` only when you need the full inline style declarations.\n\n## Create design tokens\n\nCommands:\n\n- MCP tool: create-design-token {"tokens":"tokens.json contents"}\n\n## Update design token styles\n\nCommands:\n\n- MCP tool: update-design-token-styles {"designTokenId":"<tokenId>","updates":"styles.json contents"}\n\n## Delete design token styles\n\nCommands:\n\n- MCP tool: delete-design-token-styles {"designTokenId":"<tokenId>","deletions":"styles.json contents"}\n\n## Attach design token to instances\n\nCommands:\n\n- MCP tool: attach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n\n## Detach design token from instances\n\nCommands:\n\n- MCP tool: detach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n\n## Extract design token from local styles\n\nCommands:\n\n- MCP tool: extract-design-token {"instanceIds":["<instanceId>"],"name":"Brand Primary","removeLocalProps":["color"]}\n\n## List CSS variables\n\nCommands:\n\n- MCP tool: list-css-variables {"withUsage":true}\n\n## Define CSS variables\n\nCommands:\n\n- MCP tool: define-css-variable {"vars":{"--color-primary":"#2d3748","--color-accent":"#e53e3e","--space-card":"1.5rem"},"overwrite":true}\n\nNotes:\n\n- Define or overwrite multiple CSS variables atomically by including every name and value in one `vars` object.\n- Pass colors as CSS strings. Structured `hex` color components use normalized values from `0` to `1`, not `0` to `255`.\n\n## Delete CSS variables\n\nCommands:\n\n- MCP tool: delete-css-variable {"names":["--color-primary","--color-accent","--space-card"],"force":true}\n\nNotes:\n\n- Delete multiple CSS variables atomically by including every name in one `names` array. Destructive MCP calls still require the returned confirmation token before they commit.\n\n## Rewrite CSS variable references\n\nCommands:\n\n- MCP tool: rewrite-css-variable-refs {"map":"variables.json contents"}\n\n## List data variables\n\nCommands:\n\n- MCP tool: list-variables {}\n- MCP tool: list-variables {"scopeInstanceId":"<instanceId>"}\n\nNotes:\n\n- Data variables live in the internal `dataSources` namespace.\n- For raw `snapshot`, request the public `variables` namespace rather than the internal `dataSources` name. Raw patch payloads still use `dataSources` when applying direct changes.\n- Scope variables to the instance where they should become available. Descendants can use them in expressions, and nested variables with the same name mask outer variables.\n\n## Create data variable\n\nCommands:\n\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"title","value":{"type":"string","value":"Hello"}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"count","value":{"type":"number","value":3}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"featured","value":{"type":"boolean","value":true}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"tags","value":{"type":"json","value":["news","product"]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"filters","value":{"type":"json","value":{"tag":"news"}}}\n\nNotes:\n\n- Data variable values support `string`, `number`, `boolean`, and `json`. Use `json` for all arrays and objects.\n- Parameters are internal scoped runtime values provided by pages, collections, or components. They are not a public authoring surface: do not create, update, or delete parameter records. Use data variables/resources for user-authored data, and reference documented context values such as `system` only where they are already in scope.\n\n## Update data variable\n\nCommands:\n\n- MCP tool: update-variable {"dataSourceId":"<variableId>","values":{"value":{"type":"json","value":{"count":1}}}}\n\n## Delete data variable\n\nCommands:\n\n- MCP tool: delete-variable {"dataSourceId":"<variableId>"}\n\n## List resources\n\nCommands:\n\n- MCP tool: list-resources {}\n- MCP tool: list-resources {"scopeInstanceId":"<instanceId>"}\n\n## Create resource\n\nCommands:\n\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"https://api.example.com/posts","headers":[]}}\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"\'https://api.example.com/posts?tag=\' + filters.tag","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Filtered Posts","method":"get","url":"https://api.example.com/posts","searchParams":[{"name":"tag","value":"filters.tag"},{"name":"source","value":{"type":"literal","value":"website"}}],"headers":[{"name":"Authorization","value":"\'Bearer \' + auth.token"}]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Post GraphQL","control":"graphql","method":"post","url":"https://api.example.com/graphql","headers":[{"name":"Content-Type","value":{"type":"literal","value":"application/json"}}],"body":"{ query: \'query Post($slug: String!) { post(slug: $slug) { title } }\', variables: { slug: system.params.slug } }"},"scopeInstanceId":"<instanceId>","dataSourceName":"post"}\n- MCP tool: create-resource {"resource":{"name":"Current Date","control":"system","method":"get","url":"/$resources/current-date","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"currentDate"}\n\nNotes:\n\n- Resource `url` accepts plain fixed URLs and paths such as `https://api.example.com/posts` and `/$resources/current-date`.\n- Resource `url` can also be a JavaScript expression when it is computed, such as `"https://api.example.com/posts?tag=" + filters.tag`.\n- Header values, search parameter values, and body accept expressions for dynamic values. For fixed text, use `{"type":"literal","value":"application/json"}`; Webstudio stores the required string expression for you.\n- Search parameter values, header values, and body expressions can read scoped variables and documented runtime context values such as `system` when they are available at the resource scope.\n- Add `scopeInstanceId` and `dataSourceName` when the resource result should be exposed as a scoped read data variable. Scoped resources are generated into the page resource `data` map and may be loaded during page rendering. Use this for read-oriented resources such as GET CMS/API data.\n- For submit/write/action resources, create the resource without `scopeInstanceId`, then bind a component prop such as a Form `action` with `bind-props` and `binding.type: "resource"`. Prop-bound resources are generated into the page resource `action` map instead of the read `data` map. Use this for POST, PUT, DELETE, webhooks, GraphQL submissions, and other explicit action flows.\n- Resource `method` can be `get`, `post`, `put`, or `delete`. Use GET for read data, POST for creates/GraphQL/webhooks/form submissions, PUT for full updates or replacements, and DELETE for deletion actions.\n- Optional `control` values are `graphql` and `system`. Use `graphql` for GraphQL-style requests, usually POST with a query body. Use `system` for built-in resources such as `"/$resources/sitemap.xml"`, `"/$resources/current-date"`, and `"/$resources/assets"` and when the resource should use the built-in `system` parameter. System fields are `system.origin`, `system.pathname`, `system.params`, and `system.search`.\n\n## Update resource\n\nCommands:\n\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: replace-resource-text {"find":"api.old.example.com","replace":"api.example.com","fields":["url"],"limit":20}\n\n## Query Markdown assets\n\nCommands:\n\n- MCP tool: get-asset-field-catalog {}\n- MCP tool: validate-asset-query {"query":{"where":{"all":[{"field":["extension"],"operator":"eq","value":"md"},{"field":["properties","draft"],"operator":"ne","value":true}]},"limit":20}}\n- MCP tool: create-assets-resource {"name":"All assets","scopeInstanceId":"<instanceId>","dataSourceName":"assets"}\n- MCP tool: create-assets-resource {"name":"Published posts","scopeInstanceId":"<instanceId>","dataSourceName":"posts","query":{"result":"many","where":{"all":[{"field":["extension"],"operator":"eq","value":{"type":"literal","value":"md"}},{"field":["properties","draft"],"operator":"ne","value":{"type":"literal","value":true}}]},"sort":[{"field":["properties","publishedAt"],"direction":"desc"}],"limit":{"type":"literal","value":20},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"],["properties","slug"],["properties","publishedAt"],["properties","excerpt"]]},"content":{"mode":"none"}}}\n- MCP tool: create-assets-resource {"name":"Post by slug or ID","scopeInstanceId":"<instanceId>","dataSourceName":"post","query":{"result":"one","where":{"all":[{"field":["extension"],"operator":"eq","value":{"type":"literal","value":"md"}},{"field":["properties","slug"],"operator":"eq","value":"system.params.slug"}]},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"],["properties","publishedAt"]]},"content":{"mode":"markdown-body-ref"}}}\n- MCP tool: list-assets-resources {}\n- MCP tool: get-assets-resource {"resourceId":"<resourceId>"}\n- MCP tool: update-assets-resource {"resourceId":"<resourceId>","values":{"query":null}}\n- MCP tool: preview-asset-query {"query":{"result":"one","where":{"all":[{"field":["extension"],"operator":"eq","value":"md"},{"field":["properties","slug"],"operator":"eq","value":"hello-world"}]},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"]]},"content":{"mode":"markdown-body-ref"}}}\n\nNotes:\n\n- Read the field catalog before authoring unfamiliar queries. It includes dynamic schema-less frontmatter paths such as `properties.author.name`, observed types, optionality, and mixed-type state without downloading the source files.\n- Minimize the deployed content database by using `output.mode:"fields"` and selecting only fields the rendered page needs. Keep `includeMetadata:false` unless the rendered value needs file metadata such as name, path, MIME type, or creation date, and avoid `output.mode:"all"` as a convenience default. Query diagnostics are returned separately and do not require metadata output. Filters and sorting may still require their referenced fields in the database.\n- Every reachable Assets data source contributes to the shared published database. Keep one final resource per rendered query. Update an existing scoped resource rather than creating a placeholder, preview copy, or repair replacement, and remove obsolete duplicates.\n- Keep bounded overview filters, limits, and offsets literal and add a deterministic ID tie-breaker to the sort. With `content.mode:"none"`, compilation can materialize the small overview result instead of retaining its output fields across every candidate document. Use runtime expressions only for genuinely dynamic values such as the detail slug.\n- Combine filters with `where.all` (AND) and `where.any` (OR), including nested groups. Filter values, limit, and offset on a saved resource may be Webstudio expressions evaluated at render time. Preview queries use concrete JSON values.\n- Query Markdown or MDX files directly and use `content.mode:"markdown-body-ref"` when rendering their bodies. The published database retains metadata and a document reference, then fetches only the selected files from Asset storage at runtime; it does not embed their bodies. Filter and paginate before content is loaded.\n- `full` and bounded `range` continue to request embedded file bytes. Use them only when the caller requires the complete source or a byte range.\n- In Markdown or MDX, reference sibling Assets with conventional relative URLs such as `../images/hero.png`. Deferred `markdown-body-ref` content resolves matching files against the document\'s folder and emits the correct Builder or published Asset URL. Keep external URLs absolute.\n- Markdown Embed renders sanitized authored HTML for figures, captions, audio, video, and iframes. Scripts, inline event handlers, `srcdoc`, and unsafe URL protocols are removed. It does not render Webstudio MDX elements; connect the `.mdx` file to a Content Block for that workflow.\n- Preview each query with concrete values before saving it. Inspect `__diagnostics__.query` for the temporary query-only footprint and `__diagnostics__.database` for the merged database built from all reachable Assets queries. Only `database.usedBytes` counts toward `database.maxBytes`; query sizes must not be summed and are not separate allowances. Compare `usedBytes`, `unboundedBytes`, and `truncated` within both scopes. A completed Markdown blog should include every source document without truncation, contain no embedded Markdown bodies, and retain only its intended materialized overview query. When merged usage approaches the limit, remove duplicate reachable resources first, then unused output fields, then narrow candidate files.\n- Use `result:"many"` for listings. It returns the ID-keyed map at `<dataSourceName>.data`, with `totalCount` and `hasMore` at `<dataSourceName>.meta`, and remains the default for existing queries. Bind a listing Collection to `posts.data`.\n- Use `result:"one"` for a unique detail route. It returns the selected item or `null` directly at `post.data`, always includes its `id`, and omits pagination. Bind components and page settings directly with expressions such as `post.data.properties.title`, `post.data.content.text`, and `post.data ? 200 : 404`. `result:"first"` and `result:"last"` also return a direct item or `null` but require explicit sorting.\n- Assets always executes a structured query. Omit `query` to use the default many-result query, which selects URL and optional image dimensions. Set `values.query:null` to restore it.\n- `create-assets-resource` and `update-assets-resource` are the semantic authoring path. Do not construct the internal query URL, headers, or body expression manually.\n- The shared metadata index is maintained automatically and is emitted only when a configured Assets resource is reachable.\n\n## Delete resource\n\nCommands:\n\n- MCP tool: delete-resource {"resourceId":"<resourceId>"}\n\n## List assets\n\nCommands:\n\n- MCP tool: list-assets {"withUsage":true}\n- MCP tool: list-assets {"verbose":true}\n\nNotes:\n\n- Compact results include each asset\'s folder id. Use `verbose:true` to include complete records for a page of assets, or `get-asset` to read one complete record including description, folder, creation time, and image/font metadata.\n- Image asset descriptions are the default alt text for asset-backed Image components.\n- To generate missing descriptions, inspect the image in its rendered page or asset source, write a concise description of its purpose, and save it on the asset rather than duplicating it on each Image instance.\n\n## Get asset\n\nCommands:\n\n- MCP tool: get-asset {"assetId":"<assetId>"}\n\n## List asset folders\n\nCommands:\n\n- MCP tool: list-asset-folders {}\n\n## Create asset folder\n\nCommands:\n\n- MCP tool: create-asset-folder {"name":"Marketing"}\n- MCP tool: create-asset-folder {"name":"Photos","parentId":"<parentFolderId>"}\n\n## Update asset folder\n\nCommands:\n\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"name":"Brand"}}\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"parentId":"<parentFolderId>"}}\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"parentId":null}}\n\nNotes:\n\n- Updating `parentId` is the folder equivalent of cut and paste. Use `null` to move a folder to Root.\n\n## Duplicate asset folder\n\nCommands:\n\n- MCP tool: duplicate-asset-folder {"folderId":"<folderId>"}\n- MCP tool: duplicate-asset-folder {"folderId":"<folderId>","parentId":"<targetFolderId>"}\n\nNotes:\n\n- Duplication recursively copies descendant folders and assets. This is the folder equivalent of copy and paste.\n\n## Delete asset folder\n\nCommands:\n\n- MCP tool: delete-asset-folder {"folderId":"<folderId>"}\n\nNotes:\n\n- Deleting a folder recursively deletes its descendant folders and assets.\n\n## Update asset metadata\n\nCommands:\n\n- MCP tool: update-asset {"assetId":"<assetId>","values":{"description":"Team collaborating around a whiteboard"}}\n- MCP tool: update-asset {"assetId":"<fontAssetId>","values":{"meta":{"family":"Rajdhani","style":"normal","weight":600}}}\n\nNotes:\n\n- Use an empty description only when the image is intentionally decorative.\n- Updating an image asset description updates the default alt text wherever that asset is used with an asset-backed alt prop.\n- Font metadata updates merge with the detected metadata and are validated before committing; use this to correct a family, style, or weight after upload.\n\n## Generate missing image descriptions with an agent\n\nCommands:\n\n- MCP tool: audit {"scopes":["accessibility"],"verbose":true}\n- MCP tool: set-image-descriptions {"updates":[{"assetId":"hero-id","description":"Team collaborating around a whiteboard"},{"assetId":"texture-id","decorative":true}]}\n- MCP tool: audit {"scopes":["accessibility"]}\n\nNotes:\n\n- Start from `missing-image-description` findings. Inspect each image in its rendered page context before writing text.\n- The vision-capable agent generates the wording; the CLI validates and stores it but does not contain its own vision model.\n- Use `decorative:true` only when the image adds no information. This intentionally stores an empty description so later audits do not report it as missing.\n- Re-run the accessibility audit after the update. Asset-backed Image components use the saved asset description as their default alt text.\n\n## Manage fonts\n\nCommands:\n\n- MCP tool: list-fonts {"includeSystem":true}\n- MCP tool: list-assets {"type":"font"}\n- MCP tool: upload-asset {"asset":{"name":"acme-sans.woff2","type":"font","format":"woff2","meta":{"family":"Acme Sans","style":"normal","weight":400}},"assetsDir":".webstudio/assets"}\n- MCP tool: update-styles {"updates":"styles.json contents"}\n\nNotes:\n\n- Use `list-fonts` to discover uploaded families and system stacks. Upload/delete fonts through the existing asset tools, then apply a family with a `font-family` style declaration.\n\n## Upload one asset\n\nCommands:\n\n- MCP tool: upload-asset {"asset":{"name":"image.png","type":"image","format":"png","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n- MCP tool: upload-asset {"asset":{"name":"image.png","type":"image","format":"png","folderId":"<folderId>","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n\n## Upload asset batch\n\nCommands:\n\n- MCP tool: upload-assets {"assets":[{"name":"image.png","type":"image","format":"png","meta":{"width":1200,"height":630}}],"assetsDir":".webstudio/assets"}\n\nNotes:\n\n- Multi-file uploads return `uploaded`, `failed`, and `ambiguous` lists with the\n original input index. Retry only `failed` files. A forced upload in\n `ambiguous` may already be committed; inspect the Assets list before deciding\n whether to upload it again. Successful uploads remain committed.\n\n## Duplicate asset\n\nCommands:\n\n- MCP tool: duplicate-asset {"assetId":"<assetId>"}\n- MCP tool: duplicate-asset {"assetId":"<assetId>","folderId":"<targetFolderId>"}\n- MCP tool: duplicate-asset {"assetId":"<assetId>","folderId":null}\n\nNotes:\n\n- Duplication is the asset equivalent of copy and paste. Updating `folderId` is the equivalent of cut and paste; use `null` for Root.\n\n## Download asset\n\nCommands:\n\n- MCP tool: download-asset {"assetId":"<assetId>"}\n\n## Find asset usage\n\nCommands:\n\n- MCP tool: find-asset-usage {"assetId":"<assetId>"}\n\n## Replace asset references\n\nCommands:\n\n- MCP tool: replace-asset {"fromAssetId":"<oldAssetId>","toAssetId":"<newAssetId>"}\n\n## Delete assets\n\nCommands:\n\n- MCP tool: delete-asset {"assetIdsOrPrefixes":["<assetId>"],"force":true}\n\n## Publish project\n\nCommands:\n\n- webstudio publish deploy --target production --json\n\n## List publishes\n\nCommands:\n\n- webstudio publish list --json\n\n## Check publish job\n\nCommands:\n\n- webstudio publish status --job <buildId> --json\n\n## Unpublish\n\nCommands:\n\n- webstudio publish unpublish --target production --confirm --json\n\n## List domains\n\nCommands:\n\n- webstudio domains list --json\n\n## Create domain\n\nCommands:\n\n- webstudio domains create --domain example.com --json\n\n## Update domain\n\nCommands:\n\n- webstudio domains update --domain-id <domainId> --domain www.example.com --json\n\n## Delete domain\n\nCommands:\n\n- webstudio domains delete --domain-id <domainId> --confirm --json\n\n## Verify domain\n\nCommands:\n\n- webstudio domains verify --domain-id <domainId> --json\n\n## Make arbitrary store-level changes\n\nCommands:\n\n- MCP tool: inspect {}\n- MCP tool: snapshot {"include":["<namespace>"]}\n- MCP tool: apply-patch {"baseVersion":"<version>","transactions":"patch.json contents"}\n\nNotes:\n\n- Use only when no semantic command exists.\n\n## Manage marketplace metadata\n\nCommands:\n\n- MCP tool: get-marketplace-product {}\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"asset-id","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n- MCP tool: submit-marketplace-product {"acknowledgePublicSubmission":true}\n\nPatch namespaces:\n\n- marketplaceProduct\n\n## Search and inspect safely\n\nCommands:\n\n- webstudio search-project \'{"query":"pricing"}\'\n- MCP tool: search-project {"query":"pricing"}\n- MCP tool: search-project {"query":"api.example.com","namespaces":["resources"]}\n- MCP tool: search-project {"query":"Brand","namespaces":["styles","styleSources","styleSourceSelections"]}\n- MCP tool: list-instances {"pagePath":"/","maxDepth":5}\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["props","styles","children"]}\n- MCP tool: list-texts {"pagePath":"/"}\n- MCP tool: list-assets {"withUsage":true}\n- MCP tool: find-asset-usage {"assetId":"<assetId>"}\n- MCP tool: snapshot {"include":["pages","instances","props","resources","assets"]}\n\nNotes:\n\n- `search-project` is a standalone local-session command over every Builder namespace. It returns matching values with stable match ids, namespace paths, owning entities, references, and affected pages.\n- Use it when a value or id is known. Use list/get tools when the project structure or target is not known.\n- The command synchronizes its required Builder namespaces, then searches locally and returns only matches. Namespace filters limit values matched; related namespaces may still supply route and reference context, and synchronization is unchanged. This reduces data passed to the model, not the initial project-data synchronization.\n- Asset records and file metadata are searchable. Asset binary or document contents are not Builder namespace data and are not searched.\n- Values in recognized credential fields are excluded and counted in `excludedSensitiveValueCount`.\n- Use `audit` for project health findings.\n\n## Audit project quality\n\nCommands:\n\n- webstudio audit --json\n- webstudio audit --scopes accessibility --scopes seo --json\n- webstudio audit --page-path /pricing --json\n- webstudio audit --scopes accessibility --verbose --json\n- webstudio audit --rendered --page-path /pricing --json\n- webstudio audit --rendered --route-example post=/blog/hello --json\n- webstudio audit --rendered --image-domain images.example.com --json\n- MCP tool: audit {}\n- MCP tool: audit {"scopes":["accessibility","security"],"severities":["error","warning"]}\n- MCP tool: audit {"scopes":["accessibility"],"verbose":true}\n- MCP tool: audit {"scopes":["craft"],"verbose":true}\n- MCP tool: audit {"rendered":true,"verbose":true}\n\nNotes:\n\n- With no scopes, `audit` checks accessibility, security, SEO, performance settings, unused assets, ineffective Collection styles, non-GET resources exposed as render-time data, and unused or duplicate style data.\n- Craft is opt-in and read-only. Run `audit` with `scopes:["craft"]` to detect whether the project is not using Craft, partially compatible, or compatible with the versioned Craft 1.2 profile. `profileStatuses` includes the University-doc provenance and the smallest safe next action. The audit never installs Craft or changes a non-Craft project.\n- The `performance` scope reports disabled atomic CSS generation. A rendered audit also measures broken, eager below-fold, and oversized images, browser-marked render-blocking resources, and legacy font formats.\n- Rendered image and resource metrics run only when the selected scopes include `performance`; responsive layout dimensions remain available whenever `rendered:true` is requested.\n- Compact findings include stable ids, severity, message, and location. Use `--verbose` or `{"verbose":true}` for evidence, explanation, suggested remediation, skipped-check details, and manual-check workflows.\n- `summary` counts all findings before severity filtering and pagination.\n- `contractVersion` identifies the audit response contract. Handle a new value before assuming existing fields retain the same meaning.\n- Expression-, resource-, and parameter-backed values that cannot be checked reliably appear in `skippedChecks`; they are not treated as passing or failing.\n- Page filters apply to page-owned accessibility, security, and SEO checks. Asset and style usage remain project-wide to avoid false unused findings.\n- Continue paginated results with `cursor`. Restart the audit if the project version changes.\n- Verbose skipped-check and manual-check details are included on the first findings page only; their total counts remain available on every page.\n- `manualChecks` describes responsive, hierarchy, and contrast checks that require preview screenshots and vision.\n- Focused audits return only manual checks relevant to their selected scopes.\n- In a long-lived MCP session, `{"rendered":true}` reuses preview and screenshot\n tools to capture every static page at mobile, desktop, and Builder breakpoint\n edges. Compact output reports rendered check/issue/failure counts; verbose\n output includes screenshot paths and measured layout dimensions.\n- Dynamic route templates are skipped unless `--route-example <pageId>=<path>`\n (or MCP `routeExamples`) supplies a concrete path. The path must not contain\n unresolved `:` or `*` parameters.\n- Plans above 120 captures return a short-lived confirmation token. Review the\n unchanged plan, then rerun with `--confirm-large-run` and\n `--confirmation-token`.\n- Detailed rendered evidence is stored in a versioned manifest under\n `.webstudio/audits`; compact output includes its path and screenshot count.\n- Rendered checks also report broken images, eager images below the fold, and\n image sources more than 2x their rendered dimensions in both axes, including\n Webstudio instance ids and measured dimensions when available.\n- Rendered checks include sanitized Resource Timing evidence and report\n browser-marked render-blocking resources plus legacy `.ttf`, `.otf`, and\n `.woff` fonts without applying a universal transfer-size threshold.\n- Fix findings through semantic mutation commands, then rerun `audit` to confirm their deterministic finding ids disappeared.\n\n## Verify dynamic bindings\n\nCommands:\n\n- webstudio verify-bindings --json\n- MCP tool: verify-bindings {"pagePath":"/pricing"}\n- MCP tool: verify-bindings {"instanceId":"<instanceId>","limit":50}\n\nNotes:\n\n- Statically checks persisted text expressions, expression/action/resource/parameter props, resource expressions, and page metadata.\n- Findings distinguish invalid syntax, unknown or out-of-scope variables, stale internal data-source ids, and missing resource or parameter references.\n- Page and instance filters can be combined when the instance belongs to the selected page. Continue findings with `cursor`.\n- This operation does not resolve rendered values or execute external resources. Preview representative loading, empty, error, and populated states after static findings are fixed.\n\n## Refactor targeted content\n\nCommands:\n\n- MCP tool: list-instances {"pagePath":"/"}\n- MCP tool: list-texts {"pagePath":"/"}\n- MCP tool: update-text {"instanceId":"<instanceId>","childIndex":0,"text":"Launch faster"}\n- MCP tool: replace-text {"find":"Old headline","replace":"New headline","match":"exact","pagePath":"/pricing","limit":20}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans"}}}\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: replace-asset {"fromAssetId":"<oldAssetId>","toAssetId":"<newAssetId>"}\n- MCP tool: replace-styles {"property":"color","fromValue":{"type":"keyword","value":"red"},"toValue":{"type":"keyword","value":"blue"}}\n- MCP tool: rewrite-css-variable-refs {"map":"variables.json contents"}\n\nNotes:\n\n- Use focused reads first, then mutate only matching instances, props, metadata, resource URLs, assets, or style references. Use `replace-text` for bounded literal text changes, `replace-prop-text` for bounded static prop text, `replace-resource-text` for fixed resource names/URLs, and `update-text` for one known child or expressions.\n\n## Optimize existing project\n\nCommands:\n\n- MCP tool: list-pages {}\n- MCP tool: list-folders {}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans"}}}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: list-breakpoints {}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"maxWidth":1023}}\n- MCP tool: get-styles {"instanceIds":["<instanceId>"],"includeTokens":true}\n- MCP tool: update-styles {"updates":"styles.json contents"}\n- MCP tool: attach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n- MCP tool: update-project-settings {"meta":{"siteName":"Acme"}}\n\nNotes:\n\n- Use this for SEO metadata, accessibility labels, responsive behavior, token consistency, and project settings.\n\n## Connect external data\n\nCommands:\n\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"title","value":{"type":"string","value":"Hello"}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"tags","value":{"type":"json","value":["news","product"]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"filters","value":{"type":"json","value":{"tag":"news"}}}\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"https://api.example.com/posts","searchParams":[{"name":"tag","value":"filters.tag"},{"name":"source","value":{"type":"literal","value":"website"}}],"headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Post GraphQL","control":"graphql","method":"post","url":"https://api.example.com/graphql","headers":[{"name":"Content-Type","value":{"type":"literal","value":"application/json"}}],"body":"{ query: \'query Post($slug: String!) { post(slug: $slug) { title } }\', variables: { slug: system.params.slug } }"},"scopeInstanceId":"<instanceId>","dataSourceName":"post"}\n- MCP tool: create-resource {"resource":{"name":"Current Date","control":"system","method":"get","url":"/$resources/current-date","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"currentDate"}\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.collection>{/_ collection content _/}</ws.collection>"}\n\nNotes:\n\n- Use this for CMS sections, blog listings, Ghost/headless CMS pages, n8n-style integrations, and API URLs built from variables.\n- For read data, expose GET resources as scoped data variables with `scopeInstanceId`/`dataSourceName` and read the loaded result from the resource result wrapper, usually `.data`.\n- For writes, webhooks, GraphQL submissions, and deletes, prefer unscoped resources bound to Form `action` props so they become action resources instead of auto-loaded read resources.\n- Use direct props for fixed values and prop bindings only when a prop must read a data variable, resource, action, or documented runtime context value such as `system`.\n\n## Render an array or object as repeated content\n\nCommands:\n\n- MCP tool: insert-collection {"parentInstanceId":"<instanceId>","data":{"type":"expression","value":"posts.data.items"},"itemFragment":"<ws.element ws:tag=\'article\'><ws.element ws:tag=\'h2\'>{expression`collectionItem.title`}</ws.element></ws.element>"}\n- MCP tool: inspect-instance {"instanceId":"<collectionId>","include":["props","bindings","children"]}\n\nNotes:\n\n- Use Collection whenever an array or object from a resource or data variable should render a list, grid, cards, table rows, options, tabs, or other repeated UI.\n- Pass `insert-collection` the complete array or object. Do not pass the resource response wrapper or one indexed item. External resource arrays are commonly nested under the scoped resource result\'s `data` field or deeper.\n- Pass one repeated-item Webstudio JSX root. The command creates the Collection, its private current-item/current-key parameters, the iterable binding, and descendant item bindings atomically.\n- Collection renders the item root once for every entry. Use `expression` values such as `collectionItem.title` in descendant text and props. Object iteration also exposes `collectionItemKey`.\n- Wrap multiple repeated sibling instances in one Element inside Collection.\n- For repeated Radix items such as accordion items, tabs, or menu options, bind a stable unique id or slug to every required `value` prop.\n- See the [Collection documentation](https://docs.webstudio.is/university/core-components/collection) for the equivalent Builder workflow.\n\n## Support dynamic runtime behavior\n\nCommands:\n\n- MCP tool: integrate-runtime-ui {"parentInstanceId":"<instanceId>","resources":[{"resource":{"name":"Seats","method":"get","url":"https://api.example.com/seats","headers":[]},"dataSourceName":"Seats","exposeAsDataSource":true}],"structure":{"type":"collection","data":{"type":"expression","value":"Seats.data"},"itemFragment":{"children":[{"type":"id","value":"seat"}],"instances":[{"type":"instance","id":"seat","component":"Text","children":[{"type":"expression","value":"collectionItem.label"}]}],"props":[],"dataSources":[],"resources":[],"styleSources":[],"styleSourceSelections":[],"styles":[],"breakpoints":[],"assets":[]}},"retainedBehavior":[{"instanceId":"<scriptInstanceId>","responsibility":"Seat selection behavior"}]}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n- MCP tool: create-resource {"resource":{"name":"Seats","method":"get","url":"https://api.example.com/seats","headers":[]}}\n- MCP tool: snapshot {"include":["instances","props","resources"]}\n- MCP tool: apply-patch {"baseVersion":"<version>","transactions":"patch.json contents"}\n\nNotes:\n\n- Use `integrate-runtime-ui` to create variables/resources, insert one editable fragment or Collection, and add safe data bindings in one transaction.\n- List existing script-owned responsibilities under `retainedBehavior`. The operation preserves those instances and never evaluates or accepts replacement script bodies.\n- `unsupportedConversions` records behavior that cannot be represented safely. Dry-run returns the complete transaction and the same retained/unsupported report without changing the project.\n- New actions and HtmlEmbed scripts are intentionally rejected. Create normal editable components and data bindings; keep opaque runtime behavior in existing script instances.\n\n## Build authenticated pages\n\nCommands:\n\n- MCP tool: meta.guide {"brief":"Build a Supabase-authenticated account page","taskScope":"structural-project-change","workflow":"authenticated-page"}\n- MCP tool: create-page {"name":"Account","path":"/account"}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"meta":{"auth":{"method":"basic","login":"<login>","password":"<password>"}}}}\n- MCP tool: create-resource {"resource":{"name":"Session","method":"get","url":"https://api.example.com/session","headers":[]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"user","value":{"type":"json","value":{}}}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n\nNotes:\n\n- Inspect and reuse the project\'s existing auth convention before authoring. Do\n not add a second provider or session model implicitly.\n- Model signed-out, loading, signed-in, and failed-auth states explicitly.\n- Never store credentials, service-role keys, refresh tokens, private session\n values, or authenticated response bodies in project data, command output,\n screenshots, agent instructions, or error reports. Privileged provider calls\n and authorization enforcement belong server-side.\n- Basic auth is semantic today. Provider-specific Supabase/Firebase setup still\n uses the existing resource, variable, prop, binding, and embed tools; there is\n no provider-specific installer.\n\n## Generate from design input\n\nCommands:\n\n- MCP tool: meta.guide {"brief":"Recreate this Figma design as a responsive page","taskScope":"visual-change","workflow":"design-input"}\n- MCP tool: create-page {"name":"Landing","path":"/landing"}\n- MCP tool: create-design-token {"tokens":"tokens.json contents"}\n- MCP tool: define-css-variable {"vars":"vars.json contents"}\n- MCP tool: list-breakpoints {}\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.element ws:tag=\'section\'><ws.element ws:tag=\'p\'>Section copy</ws.element></ws.element>"}\n- MCP tool: update-styles {"updates":[{"instanceId":"<instanceId>","breakpoint":"<breakpointId-from-list-breakpoints>","property":"padding-left","value":{"type":"unit","unit":"px","value":24}}]}\n- MCP tool: preview.start {"host":"127.0.0.1","port":5173}\n- MCP tool: screenshot {"path":"/landing","output":"landing-desktop.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"path":"/landing","output":"landing-mobile.png","viewport":{"width":390,"height":844},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"baseUrl":"http://127.0.0.1:5177","path":"/landing","output":"landing-desktop.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n\nNotes:\n\n- Use this after the agent can inspect the supplied design. There is no direct\n Figma, screenshot, Inception, or `design.md` import command.\n- Inspect and reuse existing variables, tokens, styles, components, assets, and\n page patterns before authoring. Build semantic editable structure rather than\n flattening the design into an image or absolute-positioned approximation.\n- Ask whether the user wants visual verification unless they explicitly requested\n it. If they decline, stop after focused reads and a static audit. If they opt\n in, verify one familiar viewport inside every distinct Builder breakpoint\n range, then run rendered audit and inspect the screenshots before completion.\n\n## Cross-project maintenance\n\nCommands:\n\n- webstudio mcp run .temp/projects.json\n- webstudio mcp run .temp/projects.json --dry-run\n- webstudio mcp run .temp/projects.json --approve-mutations --concurrency 2\n\nNotes:\n\n- Put shared `calls` and a `projects` array of independently linked project roots in the existing `mcp run` manifest. Project roots are relative to the manifest file.\n- Each project uses its own config, authentication, ProjectSession storage, checkpoint, and failure boundary. Confirmed successful calls are checkpointed. Reads can resume automatically; a mutation interrupted after dispatch is reported as ambiguous and is not replayed automatically, preventing silent duplicate writes.\n- Focus the manifest on bounded reads or audits first. Use per-call `dryRun`, global `--dry-run`, or explicitly approve committed mutations with `--approve-mutations` after reviewing the manifest.\n\n# Known CLI Gaps\n\n## Provider-specific authenticated pages\n\nMissing:\nCLI supports page basic auth and generic resources/props/embeds, but not guided Supabase/Firebase auth setup.\n\nCurrent fallback:\nCall `meta.guide` with `workflow:"authenticated-page"`, then create the\npage, resources, variables, props, bindings, and embeds with existing semantic\ntools.\n\nSuggested commands:\n\n- setup-auth-page\n\n## Generate from design input\n\nMissing:\nNo command imports Figma, screenshots, Inception output, or design.md and turns it into pages/tokens/layout.\n\nCurrent fallback:\nCall `meta.guide` with `workflow:"design-input"`, let the agent inspect the supplied\ndesign, then use semantic page, token, asset, fragment, style, preview,\nscreenshot, and audit tools. Use `apply-patch` only when no semantic operation\nfits.\n\nSuggested commands:\n\n- generate-from-design\n\n## Built-in cross-project maintenance\n\nMissing:\nPublic API and CLI intentionally operate on one configured project at a time; there is no built-in multi-project discovery or loop runner.\n\nCurrent fallback:\nRun the CLI from an external script that reconfigures one project/session at a time.\n\nSuggested commands:\n\n- none\n',+ '# CLI API Use Cases\n\n## Link/configure one project\n\nCommands:\n\n- webstudio init --link <api-share-link> --json\n\nNotes:\n\n- Writes local project id and global origin/token config.\n\n## Import synced project bundle into another project\n\nCommands:\n\n- webstudio sync\n- webstudio import --to <destination-share-link>\n- MCP tool: import {"to":"<destination-share-link>"}\n\nNotes:\n\n- Imports local `.webstudio/data.json` into the destination project.\n- Destination share link must allow build/import access.\n- Use `--skip-assets` only when asset rows and files should not be imported.\n\n## Identify current token\n\nCommands:\n\n- MCP tool: whoami {}\n\n## Check token permissions\n\nCommands:\n\n- webstudio permissions --json\n\n## Inspect project/build/version\n\nCommands:\n\n- MCP tool: inspect {}\n- MCP tool: snapshot {"include":["pages","instances","styles"]}\n\n## Discover CLI/API capabilities\n\nCommands:\n\n- webstudio schema api\n- webstudio schema mcp\n- webstudio man --json\n- webstudio man llm --json\n- MCP tool: meta.index {}\n- MCP tool: meta.guide {"brief":"Create a pricing page"}\n- MCP tool: meta.get-more-tools {"brief":"update-styles"}\n- webstudio mcp list-resources\n- webstudio mcp read-resource webstudio://project/guide\n- webstudio mcp read-resource webstudio://project/expressions\n\nNotes:\n\n- Use `webstudio schema mcp` for a compact machine-readable MCP tool overview. Add `--verbose` or use focused `meta.get-more-tools` calls only when exact input schemas are needed.\n- Use focused MCP tools for discovery first: `meta.index`, `meta.guide`, `meta.get-more-tools`, `components.list`, `components.summary`, `components.search`, `components.get`, `templates.list`, and `templates.get`. Protocol clients can use `resources/list` and `resources/read`; shell agents can use `webstudio mcp list-resources` and `webstudio mcp read-resource <uri>`. Read longer resources such as `webstudio://project/tools` and `webstudio://project/components` only when focused tools are insufficient.\n- `components.summary` returns counts by default; request `{"detail":"components","limit":20}` for paginated entries. Registry list tools return compact metadata, while `components.get` and `templates.get` return focused full details.\n- Read `webstudio://project/expressions` before authoring unfamiliar computed text, prop bindings, resource expressions, actions, or Collection item bindings.\n- From a shell, call one MCP tool with the shortcut form `webstudio <tool> \'<json>\'`, for example `webstudio components.summary`. The explicit equivalent is `webstudio mcp single-op-call <tool> \'<json>\'`. Use `--input-file` for large payloads.\n\n## Inspect external shadcn registry items\n\nCommands:\n\n- webstudio registry inspect --source https://example.com/r/registry.json --item button --json\n- webstudio registry inspect --source ./registry.json --item dialog --json\n- webstudio registry inspect --source https://example.com/r/button.json --json\n\nNotes:\n\n- Reads a local or remote registry item without installing files or changing the configured Webstudio project.\n- Returns the item name, description, package and registry dependencies, file paths/targets, available docs, and a read-only compatibility report.\n- The report explicitly says whether installation or editable-component conversion is supported, lists declared requirements and manual steps, and says when arbitrary source code was not analyzed.\n- This is an inspection step only. It does not install files or change the configured project.\n\n## Understand what MCP can do\n\nCommands:\n\n- webstudio man mcp\n- MCP tool: meta.index {}\n- MCP tool: meta.guide {"brief":"What can Webstudio MCP do?"}\n\nMCP lets agents work on one configured Webstudio project. Agents can:\n\n- Inspect the linked project, token permissions, and latest editable build.\n- Read selected project data for audits, migrations, and repair.\n- Search values and ids across every Builder namespace without sending full project data to the model.\n- Audit accessibility, security, SEO, performance settings, unused assets, ineffective Collection styles, and unused or duplicate style data.\n- Create and edit pages, folders, redirects, breakpoints, and page templates.\n- Create pages from reusable templates.\n- Update page metadata, SEO fields, auth settings, and marketplace metadata.\n- Insert components and styled JSX sections.\n- Create data-driven lists, grids, cards, and similar repeated UI from array or object data in one Collection operation.\n- Move, copy, wrap, unwrap, convert, rename, retag, and delete elements.\n- Update text, rich text, props, bindings, and actions.\n- Create and update local styles, design tokens, style sources, and CSS variables.\n- Create static data variables and JSON variables.\n- Create HTTP, GraphQL, and system resources.\n- Use system resources for sitemap, current date, and assets.\n- Bind resources to rendered data or form/action props.\n- Manage nested asset folders and upload, inspect, move, duplicate, download, replace, delete, and inspect usage for assets.\n- Publish, unpublish, inspect publish jobs, and manage custom domains.\n- Start preview, capture screenshots, compare screenshot diffs, and use OCR when installed.\n\n## Inspect and refresh MCP session cache\n\nCommands:\n\n- MCP tool: status {}\n- MCP tool: status {"verbose":true}\n- MCP tool: refresh {"namespaces":["pages","instances","styles"]}\n- MCP tool: reset-session {}\n\nNotes:\n\n- Use status before a task to understand the cached ProjectSession state.\n- Use status with `{"verbose":true}` only when debugging full namespaces, freshness, compatibility, or diagnostics.\n- Use refresh when project data may have changed outside the current MCP session.\n- Use reset-session when local cached state is corrupt or incompatible.\n\n## Visually verify rendered work with AI vision\n\nCommands:\n\n- MCP tool: preview.start {"host":"127.0.0.1","port":5173}\n- MCP tool: preview.status {}\n- MCP tool: preview.stop {}\n- MCP tool: screenshot {"path":"/","output":".webstudio/screenshots/home-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"path":"/pricing","output":".webstudio/screenshots/pricing-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"baseUrl":"http://127.0.0.1:5177","path":"/pricing","output":".webstudio/screenshots/pricing-current.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff"}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff","expectedText":["Pricing","Start free"]}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/home-before.png","currentPath":".webstudio/screenshots/home-current.png","outputDir":".webstudio/screenshots/diff","expectedVisual":{"maxMismatchPercentage":2,"maxChangedRegions":3,"dominantColorChange":{"channel":"luminance","direction":"increase","minMagnitude":10}}}\n- MCP tool: screenshot.diff {"baselinePath":".webstudio/screenshots/pricing-before.png","currentPath":".webstudio/screenshots/pricing-current.png","outputDir":".webstudio/screenshots/diff"}\n- MCP tool: vision.install-ocr {"confirm":true}\n\nNotes:\n\n- Enter this workflow only when the user explicitly requests visual verification or opts in after being asked. Do not start preview, screenshots, diffs, OCR, or rendered audits automatically after a mutation.\n- `preview.status` reports whether generated output is `stale`. When no preview is running, `url`, `pid`, and `mode` are omitted. When present, `renderedProjectVersion` is the last project version materialized into the preview.\n- A managed `screenshot` or another `preview.start` refreshes stale generated output before capture.\n\n- After opt-in, use this so a vision-capable AI can see the generated site from the current MCP session. Use `path`; never pass a Webstudio Builder/share URL or capture Builder chrome.\n- For multi-page work, capture every changed page by `path` through the same preview server; no click navigation is required.\n- Iterative mode is the default: after MCP mutations, path screenshots ensure generated project files are current, wait for the exact session version, and perform an ordinary page reload without Vite HMR. The preview server and browser stay alive. Use `{"mode":"production"}` only for release-like verification; rendered audit does this automatically.\n- Calling `preview.start` after a committed mutation restarts a stale iterative server so external browsers and HTTP clients receive the newly generated project.\n- Do not call `preview.start` through one-shot `webstudio mcp single-op-call`: it is long-lived. From a shell, use `webstudio mcp run` with preview.start, screenshot, and preview.stop in one shared process, or use a real long-running MCP client.\n- From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running preview/site without generating, building, starting, or restarting preview.\n- Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview. A separate one-shot `single-op-call` process does not own another process\'s preview controller.\n- Use waitForSelector when the rendered app has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.\n- The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.\n- Preview installs generated app dependencies under `.webstudio/preview` and reuses them across regenerations.\n- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.\n- When launcher metadata is available, Preview reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.\n- Unless `npm_config_cache` is already configured, npm uses the writable `.webstudio/preview/.npm-cache` directory.\n- For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.\n- If dependency installation fails, check the reported package-manager path and network configuration, then reinstall or update the Webstudio CLI if the problem persists.\n- When a baseline exists, use screenshot.diff once per baseline/current page or viewport pair to get changed regions, OCR textAnalysis, and diff artifact paths before deciding whether the result matches. Pass expectedText for explicit pass/fail current-screen text assertions with found and missing text. Pass expectedVisual for pass/fail limits on pixel mismatch percentage, changed-region count, or the overall dominant color/brightness direction.\n- If screenshot.diff reports OCR unavailable and the user agrees to install it, call vision.install-ocr {"confirm":true}; otherwise continue with pixel diff and visual inspection.\n- Compare the PNG, OCR text evidence, and diff artifacts against the user\'s intent for layout, typography, colors, spacing, imagery, and responsive framing; then iterate with focused mutations.\n- Root CLI equivalent: `webstudio screenshot --path /pricing --output pricing.png` generates a temporary production preview, captures that route, and stops the server. For repeated captures, keep `webstudio preview` running and pass its absolute URL to `webstudio screenshot`.\n\n## List pages\n\nCommands:\n\n- MCP tool: list-pages {}\n- MCP tool: list-folders {}\n\n## Read page by id\n\nCommands:\n\n- MCP tool: get-page {"pageId":"<pageId>"}\n\n## Read page by path\n\nCommands:\n\n- MCP tool: get-page-by-path {"path":"/pricing"}\n\n## Create page\n\nCommands:\n\n- MCP tool: create-page {"name":"Pricing","path":"/pricing"}\n- MCP tool: create-page {"name":"Pricing","path":"/pricing","title":"Pricing","meta":{"description":"Plans for teams"}}\n\nNotes:\n\n- `name`, `path`, page `title`, and metadata text fields accept plain fixed values.\n- For computed page titles or metadata, send JavaScript expression code such as `pageTitle ?? "Pricing"`.\n\n## Update page settings/metadata\n\nCommands:\n\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans","status":200}}}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"meta":{"auth":{"method":"basic","login":"<login>","password":"<password>"}}}}\n\nNotes:\n\n- Page `title` and metadata text fields accept plain fixed values.\n- For computed page titles or metadata, send JavaScript expression code such as `pageTitle ?? "Pricing"`.\n- Page `status` accepts a fixed HTTP status code as a number from 200 through 599 or a JavaScript expression string for a dynamic status.\n\n## Read project settings\n\nCommands:\n\n- MCP tool: get-project-settings {}\n\nNotes:\n\n- Read `meta.agentInstructions` before making project changes. It contains the project\'s own guidance for AI agents.\n- Agent instructions are shared project guidance. Do not store credentials or other secrets there.\n\n## Update project settings\n\nCommands:\n\n- MCP tool: update-project-settings {"meta":{"siteName":"Acme"}}\n- MCP tool: update-project-settings {"meta":{"agentInstructions":"Use existing design tokens and keep product copy concise."}}\n\n## Read marketplace product\n\nCommands:\n\n- MCP tool: get-marketplace-product {}\n\n## Update marketplace product\n\nCommands:\n\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"asset-id","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n\n## Submit marketplace product\n\nCommands:\n\n- MCP tool: upload-asset {"asset":{"name":"marketplace-thumbnail.png","type":"image","format":"png","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"<uploadedAssetId>","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n- MCP tool: publish {"target":"production"}\n- MCP tool: submit-marketplace-product {"acknowledgePublicSubmission":true}\n\nNotes:\n\n- Wait for the production publish to complete before submitting the product for review.\n- Submission requires complete, valid marketplace metadata.\n\n## List redirects\n\nCommands:\n\n- MCP tool: list-redirects {}\n\n## Create redirect\n\nCommands:\n\n- MCP tool: create-redirect {"old":"/old","new":"/new","status":301}\n\n## Update redirect\n\nCommands:\n\n- MCP tool: update-redirect {"old":"/old","values":{"new":"/newer","status":302}}\n- MCP tool: update-redirect {"old":"/old","values":{"status":null}}\n\n## Delete redirect\n\nCommands:\n\n- MCP tool: delete-redirect {"old":"/old"}\n\n## Set redirects\n\nCommands:\n\n- MCP tool: set-redirects {"redirects":[{"old":"/old","new":"/new","status":"301"}]}\n\n## List breakpoints\n\nCommands:\n\n- MCP tool: list-breakpoints {}\n\n## Create breakpoint\n\nCommands:\n\n- MCP tool: create-breakpoint {"label":"Tablet","maxWidth":991}\n\n## Update breakpoint\n\nCommands:\n\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"label":"Tablet","maxWidth":1023}}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"condition":null,"minWidth":768}}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"minWidth":null,"maxWidth":null,"condition":"(hover: hover)"}}\n\n## Delete breakpoint\n\nCommands:\n\n- MCP tool: delete-breakpoint {"breakpointId":"tablet"}\n\n## Duplicate page\n\nCommands:\n\n- MCP tool: duplicate-page {"pageId":"<pageId>","name":"Pricing Copy","path":"/pricing-copy"}\n- MCP tool: duplicate-page {"pageId":"<pageId>","name":"Paris","path":"/paris","substitutions":{"text":{"London":"Paris"},"variables":{"city":{"type":"string","value":"Paris"}}}}\n- webstudio duplicate-page --page <pageId> --name Paris --path /paris --substitutions \'{"text":{"London":"Paris"},"variables":{"city":{"type":"string","value":"Paris"}}}\' --json\n\nNotes:\n\n- Text substitutions replace exact fixed text only in the duplicated page\'s text children, string props, title, and metadata.\n- Variable substitutions are keyed by copied source-variable name and use typed variable values. The operation rejects missing or ambiguous names without committing a partial duplicate.\n- Existing expressions and cloned variable/resource references keep their remapped ids.\n\n## List page templates\n\nCommands:\n\n- MCP tool: list-page-templates {}\n\n## Create page template\n\nCommands:\n\n- MCP tool: create-page-template {"name":"Landing Template","title":"Landing"}\n\n## Update page template\n\nCommands:\n\n- MCP tool: update-page-template {"templateId":"<templateId>","values":{"name":"Article Template","meta":{"description":"Reusable article layout"}}}\n\n## Delete page template\n\nCommands:\n\n- MCP tool: delete-page-template {"templateId":"<templateId>"}\n\n## Duplicate page template\n\nCommands:\n\n- MCP tool: duplicate-page-template {"templateId":"<templateId>"}\n\n## Reorder page template\n\nCommands:\n\n- MCP tool: reorder-page-template {"sourceTemplateId":"<sourceTemplateId>","targetTemplateId":"<targetTemplateId>","position":"before"}\n\n## Create page from template\n\nCommands:\n\n- MCP tool: create-page-from-template {"templateId":"<templateId>","name":"Landing","path":"/landing"}\n\n## Delete page\n\nCommands:\n\n- MCP tool: delete-page {"pageId":"<pageId>"}\n\n## List folders\n\nCommands:\n\n- MCP tool: list-folders {}\n- MCP tool: list-pages {}\n\n## Create folder\n\nCommands:\n\n- MCP tool: create-folder {"name":"Blog","slug":"blog"}\n\n## Update folder\n\nCommands:\n\n- MCP tool: update-folder {"folderId":"<folderId>","values":{"name":"Blog","slug":"blog"}}\n\n## Delete folder\n\nCommands:\n\n- MCP tool: delete-folder {"folderId":"<folderId>"}\n\n## List element instances\n\nCommands:\n\n- MCP tool: list-instances {"pagePath":"/","maxDepth":3}\n\n## Inspect one element instance\n\nCommands:\n\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["props","styles","children"]}\n\n## Insert authored JSX or one component template\n\nCommands:\n\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.element ws:tag=\'section\' ws:style={css`padding: 32px;`}><ws.element ws:tag=\'h2\'>Product OS</ws.element><radix.Switch><radix.SwitchThumb /></radix.Switch></ws.element>"}\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"@webstudio-is/sdk-components-react-radix:Switch"}\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"Form"}\n\nNotes:\n\n- Use MCP `insert-fragment` as the default way to author styled component trees. It converts JSX to a structured fragment before mutation.\n- Use only exact component ids returned by `components.search`, `components.get`, or `templates.get`. Never derive or guess component ids.\n- The `ws:` namespace contains specific Webstudio core components; it is not HTML-tag shorthand. Use `<ws.element ws:tag="div">` for a native `div` and `<ws.element ws:tag="form">` for a native form, never `<ws.div>` or `<ws.form>`.\n- For Webstudio\'s complete form structure, discover the Form component and insert its automatic template with `insert-component` using component `"Form"`.\n- MCP receives JSX as a JSON string because MCP arguments are JSON. The CLI converts it locally before the runtime mutation, so the project session receives structured Webstudio data, not JSX source.\n- In `insert-fragment` JSX, use ``ws:style={css`...`}`` for Webstudio-native CSS, or use React-style object syntax such as `style={{ padding: 24 }}` when that is simpler. Both forms create editable Webstudio style data.\n- Do not access host globals or dynamic code APIs in JSX fragments, including `process`, `globalThis`, `eval`, `Function`, or `constructor`.\n- Use Webstudio prop names such as `class` and `for`; do not use React aliases `className` or `htmlFor`.\n- Use Webstudio actions for event/action props, for example `onClick={new ActionValue(["event"], expression\\`console.log(event)\\`)}`. Do not pass JavaScript functions such as `onClick={() => ...}`.\n- Plain prop values must be JSON-compatible: `null`, strings, booleans, finite numbers, arrays, and plain objects. Do not pass `undefined`, `Symbol`, `BigInt`, `NaN`, `Infinity`, `Date`, `Map`, `Set`, class instances, or circular objects; omit the prop, use plain data, or use `expression`/`ActionValue` when the value is dynamic.\n- Template-backed components used in JSX must include required child/part components explicitly under the same parent structure as the template, for example `<radix.Switch><radix.SwitchThumb /></radix.Switch>`.\n- Webstudio applies a registered template automatically when using `insert-component`, so composed components such as Switch include required child parts and styles.\n- Use `components.list`, `components.summary`, `components.search`, `components.get`, `templates.list`, and `templates.get` to discover known registry items, component ids, props, templates, insertability, and content model. Read `webstudio://project/components` only when those focused tools are insufficient.\n- Component/template registry items use a shadcn-compatible top-level shape plus Webstudio-specific superset metadata in `meta`. They are for Builder/MCP discovery, not a published shadcn install registry yet.\n- Known components with `contentModel.category: "none"` are not standalone-insertable; insert their root component template instead so required providers/parents are included.\n- Unknown custom component ids are a low-level extension mechanism, not a discovery fallback. Agents must not synthesize them.\n\n## Make a region editable in Content mode\n\nCommands:\n\n- MCP tool: insert-component {"parentInstanceId":"<instanceId>","component":"ws:block"}\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["children"]}\n\nNotes:\n\n- When a page will be handed to a Content-mode editor, wrap every region they should be able to edit in a Content Block (`ws:block`). Content-mode editors can edit text and supported props only in Content Block descendants. Content outside those blocks remains read-only, even when it looks like ordinary editable text.\n- Put reusable insertable options inside the Content Block\'s `ws:block-template` child. A template is source material, not editor content: editors cannot edit or delete it directly. When an editor inserts a template, its copy becomes a direct child of the Content Block and is editable.\n- Before handing off a page, verify with `inspect-instance` that the intended text, images, and links are inside a Content Block, and that templates include all required styling because Content-mode editors cannot use the Style panel.\n\n## Store Content Block content in MDX\n\nCommands:\n\n- MCP tool: list-assets {"type":"file"}\n- MCP tool: upload-asset {"asset":{"name":"article.mdx","type":"file","format":"mdx","meta":{}},"assetsDir":".webstudio/assets"}\n- MCP tool: connect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":{"type":"asset","assetId":"<mdxAssetId>"}}\n- MCP tool: switch-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":{"type":"asset","assetId":"<otherMdxAssetId>"}}\n- MCP tool: inspect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n- MCP tool: edit-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","source":"# Article title\\n\\nArticle body."}\n- MCP tool: update-content-block-frontmatter {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example","properties":{"title":"Article title","draft":false}}\n- MCP tool: reload-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n- MCP tool: migrate-content-block-template-references {"assetIds":["<mdxAssetId>"],"migration":{"type":"rename","from":"Old template name","to":"New template name"}}\n- MCP tool: disconnect-content-block-source {"blockInstanceId":"<contentBlockInstanceId>","renderScope":"page:/articles/example"}\n\nNotes:\n\n- Create the local `.mdx` file under `.webstudio/assets` before calling `upload-asset`. Use the returned Asset ID when connecting it.\n- Use the Content Block source operations for connecting, switching, inspecting, editing, reloading, and disconnecting. Do not create or delete the `src` prop with generic prop tools; the source operations preserve the Body outlet and Content Block lifecycle.\n- `renderScope` is a stable key for the rendered occurrence. Use a page-based key for a direct Content Block. For a repeated Collection occurrence, use a distinct key and pass the scoped values required to resolve an expression source. For example, use `source:{"type":"expression","value":"post.assetId"}` with `variables:{"post":{"assetId":"<mdxAssetId>"}}`.\n- Connecting replaces existing Body content. If the result returns `requiresConfirmation:true`, report that replacement to the user and repeat the same call with `confirmReplacement:true` only after approval.\n- Prefer Markdown whenever it can represent the component and all authored properties. Use `<ws.element ws:tag="tag">` for a standard HTML element with authored properties that Markdown cannot express. Use `<ws.element ws:name="Template name">` only for a uniquely named top-level Content Block template.\n- Read the Content Block\'s Templates children before writing `ws:name`. The value must exactly match a unique top-level template instance name. Preserve unresolved names and report their diagnostics instead of deleting them.\n- When a template is renamed or deleted, use `migrate-content-block-template-references` to update a selected set of affected MDX files. The first call returns a plan. Report its changed-file, update, omission, and diagnostic counts, then repeat the exact request with its `confirmationToken` only after approval. A rename changes matching `ws:name` values; a removal deletes matching MDX elements. Invalid files remain unchanged and are reported in diagnostics.\n- `edit-content-block-source` replaces the complete MDX source. Preserve frontmatter and unrelated source when making a bounded edit.\n- `update-content-block-frontmatter` replaces the complete frontmatter mapping. Inspect the current source first and include every property that must remain.\n- Use `update-content-block-frontmatter` for MCP frontmatter edits. MDX-rendered elements are not persistent instance targets for generic `bind-props` or `update-text` calls. Preserve existing `mode:"readwrite"` bindings when encountered; they are valid only for exact direct paths into the connected document\'s frontmatter. Keep computed expressions and `$ref` values read-only.\n- Inspect every returned diagnostic. Invalid MDX is saved rather than silently repaired; preserve the source, report the source range, and fix only the requested or invalid part.\n- If an edit in a long-lived MCP session reports a conflict after another client saved the Asset, call `reload-content-block-source`, inspect the latest source, reapply the requested change, and retry. One-shot CLI calls refresh before each operation and normally cannot reproduce a stale session. Never overwrite the newer revision blindly.\n\n## Move elements\n\nCommands:\n\n- MCP tool: move-instance {"moves":"moves.json contents"}\n\nNotes:\n\n- Use `position: "end"` to append an instance. Repeating this for A and then B preserves the final order A, B.\n- A numeric `insertIndex` addresses the target parent\'s children before the moved instance is removed. Use it for exact placement; do not calculate the last index to append.\n- Moves in one `moves` array are applied sequentially in array order.\n\n## Clone element subtree\n\nCommands:\n\n- MCP tool: clone-instance {"sourceInstanceId":"<instanceId>","targetParentInstanceId":"<targetParentId>"}\n\n## Delete element subtree\n\nCommands:\n\n- MCP tool: delete-instance {"instanceIds":["<instanceId>"]}\n\n## List text/expression children\n\nCommands:\n\n- MCP tool: list-texts {"pagePath":"/"}\n\n## Update text child\n\nCommands:\n\n- MCP tool: update-text {"instanceId":"<instanceId>","childIndex":0,"text":"Launch faster"}\n\n## Replace bounded literal text\n\nCommands:\n\n- MCP tool: replace-text {"find":"Start free","replace":"Get started","match":"exact","pagePath":"/pricing","limit":20}\n\nNotes:\n\n- This changes only literal text children, never expression children. Scope it to pagePath or pageId and set a limit before a broad replacement.\n\n## Replace bounded static prop text\n\nCommands:\n\n- MCP tool: replace-prop-text {"find":"old.example.com","replace":"www.example.com","match":"substring","names":["href","code"],"limit":20}\n\nNotes:\n\n- This changes only static string props such as href, alt, aria-label, title, and HTML embed code. It never changes expressions, resources, actions, assets, or other dynamic bindings. Use names or instanceIds and a limit to narrow the change.\n\n## Replace bounded resource text\n\nCommands:\n\n- MCP tool: replace-resource-text {"find":"api.old.example.com","replace":"api.example.com","fields":["url"],"limit":20}\n\nNotes:\n\n- This changes resource names and fixed URL literals only. It skips dynamic URL expressions, headers, search parameters, request bodies, and GraphQL query code.\n\n## Update props\n\nCommands:\n\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: replace-prop-text {"find":"Old label","replace":"New label","names":["aria-label","title"],"limit":20}\n\nNotes:\n\n- Use this for fixed prop values such as `aria-label`, `alt`, `id`, static `href`, and other direct string/number/boolean/json prop values.\n\n## Add JSON-LD structured data\n\nCommands:\n\n- MCP tool: components.get {"component":"JsonLd"}\n- MCP tool: insert-component {"parentInstanceId":"<headSlotInstanceId>","component":"JsonLd"}\n- MCP tool: update-props {"updates":[{"instanceId":"<jsonLdInstanceId>","name":"code","type":"string","value":"{\\"@context\\":\\"https://schema.org\\",\\"@type\\":\\"Organization\\",\\"name\\":\\"Acme\\"}"}]}\n- MCP tool: bind-props {"bindings":[{"instanceId":"<jsonLdInstanceId>","name":"code","binding":{"type":"expression","value":"({ \'@context\': \'https://schema.org\', \'@type\': \'Article\', headline: post.title })"}}]}\n- MCP tool: audit {"scopes":["seo"],"pagePath":"/"}\n\nNotes:\n\n- Prefer placing `JsonLd` inside `HeadSlot`.\n- For fixed structured data, store `code` as a JSON object or array encoded as a compact string. The Builder formats it for editing.\n- For structured data containing runtime values, use `bind-props` with an expression that evaluates directly to an object or array. Do not call `JSON.stringify` or assemble JSON with string concatenation. Webstudio stores the expression as source text, evaluates it at runtime, and the `JsonLd` component validates and serializes the resulting value.\n- The semantic prop update rejects malformed JSON and structurally invalid fixed JSON-LD with a precise JSON path.\n- The SEO audit also warns about a missing top-level `@context`, unknown or superseded Schema.org terms, properties unsupported by the supplied type, and incompatible primitive value types.\n- Schema.org vocabulary findings are warnings because custom vocabularies and extensions remain valid. Dynamic JSON-LD is marked as skipped for rendered validation.\n- Do not use bindings just to set static text.\n\n## Delete props\n\nCommands:\n\n- MCP tool: delete-props {"deletions":"props.json contents"}\n\n## Bind props to expressions/resources/actions\n\nCommands:\n\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n\nNotes:\n\n- Use this only when the prop should remain dynamic: expression, resource, action, or an existing scoped runtime context value such as `system`.\n- For a fixed string value, use `update-props` with `type:"string"` and a direct `value` instead.\n\n## Read styles\n\nCommands:\n\n- MCP tool: get-styles {"instanceIds":["<instanceId>"],"includeTokens":true}\n\n## Update local styles\n\nCommands:\n\n- MCP tool: update-styles {"updates":"styles.json contents"}\n\n## Delete local styles\n\nCommands:\n\n- MCP tool: delete-styles {"deletions":"styles.json contents"}\n\n## Replace matching style values\n\nCommands:\n\n- MCP tool: replace-styles {"property":"color","fromValue":{"type":"keyword","value":"red"},"toValue":{"type":"keyword","value":"blue"}}\n\n## List design tokens\n\nCommands:\n\n- MCP tool: list-design-tokens {}\n- MCP tool: list-design-tokens {"withUsage":true}\n- MCP tool: list-design-tokens {"verbose":true}\n\nNotes:\n\n- The default response is compact and includes token id, name, declaration count, and optional usage count.\n- Use `verbose:true` only when you need the full inline style declarations.\n\n## Create design tokens\n\nCommands:\n\n- MCP tool: create-design-token {"tokens":"tokens.json contents"}\n\n## Update design token styles\n\nCommands:\n\n- MCP tool: update-design-token-styles {"designTokenId":"<tokenId>","updates":"styles.json contents"}\n\n## Delete design token styles\n\nCommands:\n\n- MCP tool: delete-design-token-styles {"designTokenId":"<tokenId>","deletions":"styles.json contents"}\n\n## Attach design token to instances\n\nCommands:\n\n- MCP tool: attach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n\n## Detach design token from instances\n\nCommands:\n\n- MCP tool: detach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n\n## Extract design token from local styles\n\nCommands:\n\n- MCP tool: extract-design-token {"instanceIds":["<instanceId>"],"name":"Brand Primary","removeLocalProps":["color"]}\n\n## List CSS variables\n\nCommands:\n\n- MCP tool: list-css-variables {"withUsage":true}\n\n## Define CSS variables\n\nCommands:\n\n- MCP tool: define-css-variable {"vars":{"--color-primary":"#2d3748","--color-accent":"#e53e3e","--space-card":"1.5rem"},"overwrite":true}\n\nNotes:\n\n- Define or overwrite multiple CSS variables atomically by including every name and value in one `vars` object.\n- Pass colors as CSS strings. Structured `hex` color components use normalized values from `0` to `1`, not `0` to `255`.\n\n## Delete CSS variables\n\nCommands:\n\n- MCP tool: delete-css-variable {"names":["--color-primary","--color-accent","--space-card"],"force":true}\n\nNotes:\n\n- Delete multiple CSS variables atomically by including every name in one `names` array. Destructive MCP calls still require the returned confirmation token before they commit.\n\n## Rewrite CSS variable references\n\nCommands:\n\n- MCP tool: rewrite-css-variable-refs {"map":"variables.json contents"}\n\n## List data variables\n\nCommands:\n\n- MCP tool: list-variables {}\n- MCP tool: list-variables {"scopeInstanceId":"<instanceId>"}\n\nNotes:\n\n- Data variables live in the internal `dataSources` namespace.\n- For raw `snapshot`, request the public `variables` namespace rather than the internal `dataSources` name. Raw patch payloads still use `dataSources` when applying direct changes.\n- Scope variables to the instance where they should become available. Descendants can use them in expressions, and nested variables with the same name mask outer variables.\n\n## Create data variable\n\nCommands:\n\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"title","value":{"type":"string","value":"Hello"}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"count","value":{"type":"number","value":3}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"featured","value":{"type":"boolean","value":true}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"tags","value":{"type":"json","value":["news","product"]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"filters","value":{"type":"json","value":{"tag":"news"}}}\n\nNotes:\n\n- Data variable values support `string`, `number`, `boolean`, and `json`. Use `json` for all arrays and objects.\n- Parameters are internal scoped runtime values provided by pages, collections, or components. They are not a public authoring surface: do not create, update, or delete parameter records. Use data variables/resources for user-authored data, and reference documented context values such as `system` only where they are already in scope.\n\n## Update data variable\n\nCommands:\n\n- MCP tool: update-variable {"dataSourceId":"<variableId>","values":{"value":{"type":"json","value":{"count":1}}}}\n\n## Delete data variable\n\nCommands:\n\n- MCP tool: delete-variable {"dataSourceId":"<variableId>"}\n\n## List resources\n\nCommands:\n\n- MCP tool: list-resources {}\n- MCP tool: list-resources {"scopeInstanceId":"<instanceId>"}\n\n## Create resource\n\nCommands:\n\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"https://api.example.com/posts","headers":[]}}\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"\'https://api.example.com/posts?tag=\' + filters.tag","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Filtered Posts","method":"get","url":"https://api.example.com/posts","searchParams":[{"name":"tag","value":"filters.tag"},{"name":"source","value":{"type":"literal","value":"website"}}],"headers":[{"name":"Authorization","value":"\'Bearer \' + auth.token"}]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Post GraphQL","control":"graphql","method":"post","url":"https://api.example.com/graphql","headers":[{"name":"Content-Type","value":{"type":"literal","value":"application/json"}}],"body":"{ query: \'query Post($slug: String!) { post(slug: $slug) { title } }\', variables: { slug: system.params.slug } }"},"scopeInstanceId":"<instanceId>","dataSourceName":"post"}\n- MCP tool: create-resource {"resource":{"name":"Current Date","control":"system","method":"get","url":"/$resources/current-date","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"currentDate"}\n\nNotes:\n\n- Resource `url` accepts plain fixed URLs and paths such as `https://api.example.com/posts` and `/$resources/current-date`.\n- Resource `url` can also be a JavaScript expression when it is computed, such as `"https://api.example.com/posts?tag=" + filters.tag`.\n- Header values, search parameter values, and body accept expressions for dynamic values. For fixed text, use `{"type":"literal","value":"application/json"}`; Webstudio stores the required string expression for you.\n- Search parameter values, header values, and body expressions can read scoped variables and documented runtime context values such as `system` when they are available at the resource scope.\n- Add `scopeInstanceId` and `dataSourceName` when the resource result should be exposed as a scoped read data variable. Scoped resources are generated into the page resource `data` map and may be loaded during page rendering. Use this for read-oriented resources such as GET CMS/API data.\n- For submit/write/action resources, create the resource without `scopeInstanceId`, then bind a component prop such as a Form `action` with `bind-props` and `binding.type: "resource"`. Prop-bound resources are generated into the page resource `action` map instead of the read `data` map. Use this for POST, PUT, DELETE, webhooks, GraphQL submissions, and other explicit action flows.\n- Resource `method` can be `get`, `post`, `put`, or `delete`. Use GET for read data, POST for creates/GraphQL/webhooks/form submissions, PUT for full updates or replacements, and DELETE for deletion actions.\n- Optional `control` values are `graphql` and `system`. Use `graphql` for GraphQL-style requests, usually POST with a query body. Use `system` for built-in resources such as `"/$resources/sitemap.xml"`, `"/$resources/current-date"`, and `"/$resources/assets"` and when the resource should use the built-in `system` parameter. System fields are `system.origin`, `system.pathname`, `system.params`, and `system.search`.\n\n## Update resource\n\nCommands:\n\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: replace-resource-text {"find":"api.old.example.com","replace":"api.example.com","fields":["url"],"limit":20}\n\n## Query Markdown assets\n\nCommands:\n\n- MCP tool: get-asset-field-catalog {}\n- MCP tool: validate-asset-query {"query":{"where":{"all":[{"field":["extension"],"operator":"eq","value":"md"},{"field":["properties","draft"],"operator":"ne","value":true}]},"limit":20}}\n- MCP tool: create-assets-resource {"name":"All assets","scopeInstanceId":"<instanceId>","dataSourceName":"assets"}\n- MCP tool: create-assets-resource {"name":"Published posts","scopeInstanceId":"<instanceId>","dataSourceName":"posts","query":{"result":"many","where":{"all":[{"field":["extension"],"operator":"eq","value":{"type":"literal","value":"md"}},{"field":["properties","draft"],"operator":"ne","value":{"type":"literal","value":true}}]},"sort":[{"field":["properties","publishedAt"],"direction":"desc"}],"limit":{"type":"literal","value":20},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"],["properties","slug"],["properties","publishedAt"],["properties","excerpt"]]},"content":{"mode":"none"}}}\n- MCP tool: create-assets-resource {"name":"Post by slug or ID","scopeInstanceId":"<instanceId>","dataSourceName":"post","query":{"result":"one","where":{"all":[{"field":["extension"],"operator":"eq","value":{"type":"literal","value":"md"}},{"field":["properties","slug"],"operator":"eq","value":"system.params.slug"}]},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"],["properties","publishedAt"]]},"content":{"mode":"markdown-body-ref"}}}\n- MCP tool: list-assets-resources {}\n- MCP tool: get-assets-resource {"resourceId":"<resourceId>"}\n- MCP tool: update-assets-resource {"resourceId":"<resourceId>","values":{"query":null}}\n- MCP tool: preview-asset-query {"query":{"result":"one","where":{"all":[{"field":["extension"],"operator":"eq","value":"md"},{"field":["properties","slug"],"operator":"eq","value":"hello-world"}]},"output":{"mode":"fields","includeMetadata":false,"fields":[["properties","title"]]},"content":{"mode":"markdown-body-ref"}}}\n\nNotes:\n\n- Read the field catalog before authoring unfamiliar queries. It includes dynamic schema-less frontmatter paths such as `properties.author.name`, observed types, optionality, and mixed-type state without downloading the source files.\n- Minimize the deployed content database by using `output.mode:"fields"` and selecting only fields the rendered page needs. Keep `includeMetadata:false` unless the rendered value needs file metadata such as name, path, MIME type, or creation date, and avoid `output.mode:"all"` as a convenience default. Query diagnostics are returned separately and do not require metadata output. Filters and sorting may still require their referenced fields in the database.\n- Every reachable Assets data source contributes to the shared published database. Keep one final resource per rendered query. Update an existing scoped resource rather than creating a placeholder, preview copy, or repair replacement, and remove obsolete duplicates.\n- Keep bounded overview filters, limits, and offsets literal and add a deterministic ID tie-breaker to the sort. With `content.mode:"none"`, compilation can materialize the small overview result instead of retaining its output fields across every candidate document. Use runtime expressions only for genuinely dynamic values such as the detail slug.\n- Combine filters with `where.all` (AND) and `where.any` (OR), including nested groups. Filter values, limit, and offset on a saved resource may be Webstudio expressions evaluated at render time. Preview queries use concrete JSON values.\n- Query Markdown or MDX files directly and use `content.mode:"markdown-body-ref"` when rendering their bodies. The published database retains metadata and a document reference, then fetches only the selected files from Asset storage at runtime; it does not embed their bodies. Filter and paginate before content is loaded.\n- `full` and bounded `range` continue to request embedded file bytes. Use them only when the caller requires the complete source or a byte range.\n- In Markdown or MDX, reference sibling Assets with conventional relative URLs such as `../images/hero.png`. Deferred `markdown-body-ref` content resolves matching files against the document\'s folder and emits the correct Builder or published Asset URL. Keep external URLs absolute.\n- Markdown Embed renders sanitized authored HTML for figures, captions, audio, video, and iframes. Scripts, inline event handlers, `srcdoc`, and unsafe URL protocols are removed. It does not render Webstudio MDX elements; connect the `.mdx` file to a Content Block for that workflow.\n- Preview each query with concrete values before saving it. Inspect `__diagnostics__.query` for the temporary query-only footprint and `__diagnostics__.database` for the merged database built from all reachable Assets queries. Only `database.usedBytes` counts toward `database.maxBytes`; query sizes must not be summed and are not separate allowances. Compare `usedBytes`, `unboundedBytes`, and `truncated` within both scopes. A completed Markdown blog should include every source document without truncation, contain no embedded Markdown bodies, and retain only its intended materialized overview query. When merged usage approaches the limit, remove duplicate reachable resources first, then unused output fields, then narrow candidate files.\n- Use `result:"many"` for listings. It returns the ID-keyed map at `<dataSourceName>.data`, with `totalCount` and `hasMore` at `<dataSourceName>.meta`, and remains the default for existing queries. Bind a listing Collection to `posts.data`.\n- Use `result:"one"` for a unique detail route. It returns the selected item or `null` directly at `post.data`, always includes its `id`, and omits pagination. Bind components and page settings directly with expressions such as `post.data.properties.title`, `post.data.content.text`, and `post.data ? 200 : 404`. `result:"first"` and `result:"last"` also return a direct item or `null` but require explicit sorting.\n- Assets always executes a structured query. Omit `query` to use the default many-result query, which selects URL and optional image dimensions. Set `values.query:null` to restore it.\n- `create-assets-resource` and `update-assets-resource` are the semantic authoring path. Do not construct the internal query URL, headers, or body expression manually.\n- The shared metadata index is maintained automatically and is emitted only when a configured Assets resource is reachable.\n\n## Delete resource\n\nCommands:\n\n- MCP tool: delete-resource {"resourceId":"<resourceId>"}\n\n## List assets\n\nCommands:\n\n- MCP tool: list-assets {"withUsage":true}\n- MCP tool: list-assets {"verbose":true}\n\nNotes:\n\n- Compact results include each asset\'s folder id. Use `verbose:true` to include complete records for a page of assets, or `get-asset` to read one complete record including description, folder, creation time, and image/font metadata.\n- Image asset descriptions are the default alt text for asset-backed Image components.\n- To generate missing descriptions, inspect the image in its rendered page or asset source, write a concise description of its purpose, and save it on the asset rather than duplicating it on each Image instance.\n\n## Get asset\n\nCommands:\n\n- MCP tool: get-asset {"assetId":"<assetId>"}\n\n## List asset folders\n\nCommands:\n\n- MCP tool: list-asset-folders {}\n\n## Create asset folder\n\nCommands:\n\n- MCP tool: create-asset-folder {"name":"Marketing"}\n- MCP tool: create-asset-folder {"name":"Photos","parentId":"<parentFolderId>"}\n\n## Update asset folder\n\nCommands:\n\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"name":"Brand"}}\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"parentId":"<parentFolderId>"}}\n- MCP tool: update-asset-folder {"folderId":"<folderId>","values":{"parentId":null}}\n\nNotes:\n\n- Updating `parentId` is the folder equivalent of cut and paste. Use `null` to move a folder to Root.\n\n## Duplicate asset folder\n\nCommands:\n\n- MCP tool: duplicate-asset-folder {"folderId":"<folderId>"}\n- MCP tool: duplicate-asset-folder {"folderId":"<folderId>","parentId":"<targetFolderId>"}\n\nNotes:\n\n- Duplication recursively copies descendant folders and assets. This is the folder equivalent of copy and paste.\n\n## Delete asset folder\n\nCommands:\n\n- MCP tool: delete-asset-folder {"folderId":"<folderId>"}\n\nNotes:\n\n- Deleting a folder recursively deletes its descendant folders and assets.\n\n## Update asset metadata\n\nCommands:\n\n- MCP tool: update-asset {"assetId":"<assetId>","values":{"description":"Team collaborating around a whiteboard"}}\n- MCP tool: update-asset {"assetId":"<fontAssetId>","values":{"meta":{"family":"Rajdhani","style":"normal","weight":600}}}\n\nNotes:\n\n- Use an empty description only when the image is intentionally decorative.\n- Updating an image asset description updates the default alt text wherever that asset is used with an asset-backed alt prop.\n- Font metadata updates merge with the detected metadata and are validated before committing; use this to correct a family, style, or weight after upload.\n\n## Generate missing image descriptions with an agent\n\nCommands:\n\n- MCP tool: audit {"scopes":["accessibility"],"verbose":true}\n- MCP tool: set-image-descriptions {"updates":[{"assetId":"hero-id","description":"Team collaborating around a whiteboard"},{"assetId":"texture-id","decorative":true}]}\n- MCP tool: audit {"scopes":["accessibility"]}\n\nNotes:\n\n- Start from `missing-image-description` findings. Inspect each image in its rendered page context before writing text.\n- The vision-capable agent generates the wording; the CLI validates and stores it but does not contain its own vision model.\n- Use `decorative:true` only when the image adds no information. This intentionally stores an empty description so later audits do not report it as missing.\n- Re-run the accessibility audit after the update. Asset-backed Image components use the saved asset description as their default alt text.\n\n## Manage fonts\n\nCommands:\n\n- MCP tool: list-fonts {"includeSystem":true}\n- MCP tool: list-assets {"type":"font"}\n- MCP tool: upload-asset {"asset":{"name":"acme-sans.woff2","type":"font","format":"woff2","meta":{"family":"Acme Sans","style":"normal","weight":400}},"assetsDir":".webstudio/assets"}\n- MCP tool: update-styles {"updates":"styles.json contents"}\n\nNotes:\n\n- Use `list-fonts` to discover uploaded families and system stacks. Upload/delete fonts through the existing asset tools, then apply a family with a `font-family` style declaration.\n\n## Upload one asset\n\nCommands:\n\n- MCP tool: upload-asset {"asset":{"name":"image.png","type":"image","format":"png","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n- MCP tool: upload-asset {"asset":{"name":"image.png","type":"image","format":"png","folderId":"<folderId>","meta":{"width":1200,"height":630}},"assetsDir":".webstudio/assets"}\n\n## Upload asset batch\n\nCommands:\n\n- MCP tool: upload-assets {"assets":[{"name":"image.png","type":"image","format":"png","meta":{"width":1200,"height":630}}],"assetsDir":".webstudio/assets"}\n\nNotes:\n\n- Multi-file uploads return `uploaded`, `failed`, and `ambiguous` lists with the\n original input index. Retry only `failed` files. A forced upload in\n `ambiguous` may already be committed; inspect the Assets list before deciding\n whether to upload it again. Successful uploads remain committed.\n\n## Duplicate asset\n\nCommands:\n\n- MCP tool: duplicate-asset {"assetId":"<assetId>"}\n- MCP tool: duplicate-asset {"assetId":"<assetId>","folderId":"<targetFolderId>"}\n- MCP tool: duplicate-asset {"assetId":"<assetId>","folderId":null}\n\nNotes:\n\n- Duplication is the asset equivalent of copy and paste. Updating `folderId` is the equivalent of cut and paste; use `null` for Root.\n\n## Download asset\n\nCommands:\n\n- MCP tool: download-asset {"assetId":"<assetId>"}\n\n## Find asset usage\n\nCommands:\n\n- MCP tool: find-asset-usage {"assetId":"<assetId>"}\n\n## Replace asset references\n\nCommands:\n\n- MCP tool: replace-asset {"fromAssetId":"<oldAssetId>","toAssetId":"<newAssetId>"}\n\n## Delete assets\n\nCommands:\n\n- MCP tool: delete-asset {"assetIdsOrPrefixes":["<assetId>"],"force":true}\n\n## Publish project\n\nCommands:\n\n- webstudio publish deploy --target production --json\n\n## List publishes\n\nCommands:\n\n- webstudio publish list --json\n\n## Check publish job\n\nCommands:\n\n- webstudio publish status --job <buildId> --json\n\n## Unpublish\n\nCommands:\n\n- webstudio publish unpublish --target production --confirm --json\n\n## List domains\n\nCommands:\n\n- webstudio domains list --json\n\n## Create domain\n\nCommands:\n\n- webstudio domains create --domain example.com --json\n\n## Update domain\n\nCommands:\n\n- webstudio domains update --domain-id <domainId> --domain www.example.com --json\n\n## Delete domain\n\nCommands:\n\n- webstudio domains delete --domain-id <domainId> --confirm --json\n\n## Verify domain\n\nCommands:\n\n- webstudio domains verify --domain-id <domainId> --json\n\n## Make arbitrary store-level changes\n\nCommands:\n\n- MCP tool: inspect {}\n- MCP tool: snapshot {"include":["<namespace>"]}\n- MCP tool: apply-patch {"baseVersion":"<version>","transactions":"patch.json contents"}\n\nNotes:\n\n- Use only when no semantic command exists.\n\n## Manage marketplace metadata\n\nCommands:\n\n- MCP tool: get-marketplace-product {}\n- MCP tool: update-marketplace-product {"category":"pageTemplates","name":"Acme Template","thumbnailAssetId":"asset-id","author":"Acme Studio","email":"[email protected]","website":"https://example.com","issues":"","description":"Reusable template project for Acme landing pages."}\n- MCP tool: submit-marketplace-product {"acknowledgePublicSubmission":true}\n\nPatch namespaces:\n\n- marketplaceProduct\n\n## Search and inspect safely\n\nCommands:\n\n- webstudio search-project \'{"query":"pricing"}\'\n- MCP tool: search-project {"query":"pricing"}\n- MCP tool: search-project {"query":"api.example.com","namespaces":["resources"]}\n- MCP tool: search-project {"query":"Brand","namespaces":["styles","styleSources","styleSourceSelections"]}\n- MCP tool: list-instances {"pagePath":"/","maxDepth":5}\n- MCP tool: inspect-instance {"instanceId":"<instanceId>","include":["props","styles","children"]}\n- MCP tool: list-texts {"pagePath":"/"}\n- MCP tool: list-assets {"withUsage":true}\n- MCP tool: find-asset-usage {"assetId":"<assetId>"}\n- MCP tool: snapshot {"include":["pages","instances","props","resources","assets"]}\n\nNotes:\n\n- `search-project` is a standalone local-session command over every Builder namespace. It returns matching values with stable match ids, namespace paths, owning entities, references, and affected pages.\n- Use it when a value or id is known. Use list/get tools when the project structure or target is not known.\n- The command synchronizes its required Builder namespaces, then searches locally and returns only matches. Namespace filters limit values matched; related namespaces may still supply route and reference context, and synchronization is unchanged. This reduces data passed to the model, not the initial project-data synchronization.\n- Asset records and file metadata are searchable. Asset binary or document contents are not Builder namespace data and are not searched.\n- Values in recognized credential fields are excluded and counted in `excludedSensitiveValueCount`.\n- Use `audit` for project health findings.\n\n## Audit project quality\n\nCommands:\n\n- webstudio audit --json\n- webstudio audit --scopes accessibility --scopes seo --json\n- webstudio audit --page-path /pricing --json\n- webstudio audit --scopes accessibility --verbose --json\n- webstudio audit --rendered --page-path /pricing --json\n- webstudio audit --rendered --route-example post=/blog/hello --json\n- webstudio audit --rendered --image-domain images.example.com --json\n- MCP tool: audit {}\n- MCP tool: audit {"scopes":["accessibility","security"],"severities":["error","warning"]}\n- MCP tool: audit {"scopes":["accessibility"],"verbose":true}\n- MCP tool: audit {"scopes":["craft"],"verbose":true}\n- MCP tool: audit {"rendered":true,"verbose":true}\n\nNotes:\n\n- With no scopes, `audit` checks accessibility, security, SEO, performance settings, unused assets, ineffective Collection styles, non-GET resources exposed as render-time data, and unused or duplicate style data.\n- Craft is opt-in and read-only. Run `audit` with `scopes:["craft"]` to detect whether the project is not using Craft, partially compatible, or compatible with the versioned Craft 1.2 profile. `profileStatuses` includes the University-doc provenance and the smallest safe next action. The audit never installs Craft or changes a non-Craft project.\n- The `performance` scope reports disabled atomic CSS generation. A rendered audit also measures broken, eager below-fold, and oversized images, browser-marked render-blocking resources, and legacy font formats.\n- Rendered image and resource metrics run only when the selected scopes include `performance`; responsive layout dimensions remain available whenever `rendered:true` is requested.\n- Compact findings include stable ids, severity, message, and location. Use `--verbose` or `{"verbose":true}` for evidence, explanation, suggested remediation, skipped-check details, and manual-check workflows.\n- `summary` counts all findings before severity filtering and pagination.\n- `contractVersion` identifies the audit response contract. Handle a new value before assuming existing fields retain the same meaning.\n- Expression-, resource-, and parameter-backed values that cannot be checked reliably appear in `skippedChecks`; they are not treated as passing or failing.\n- Page filters apply to page-owned accessibility, security, and SEO checks. Asset and style usage remain project-wide to avoid false unused findings.\n- Continue paginated results with `cursor`. Restart the audit if the project version changes.\n- Verbose skipped-check and manual-check details are included on the first findings page only; their total counts remain available on every page.\n- `manualChecks` describes responsive, hierarchy, and contrast checks that require preview screenshots and vision.\n- Focused audits return only manual checks relevant to their selected scopes.\n- In a long-lived MCP session, `{"rendered":true}` reuses preview and screenshot\n tools to capture every static page at mobile, desktop, and Builder breakpoint\n edges. Compact output reports rendered check/issue/failure counts; verbose\n output includes screenshot paths and measured layout dimensions.\n- Dynamic route templates are skipped unless `--route-example <pageId>=<path>`\n (or MCP `routeExamples`) supplies a concrete path. The path must not contain\n unresolved `:` or `*` parameters.\n- Plans above 120 captures return a short-lived confirmation token. Review the\n unchanged plan, then rerun with `--confirm-large-run` and\n `--confirmation-token`.\n- Detailed rendered evidence is stored in a versioned manifest under\n `.webstudio/audits`; compact output includes its path and screenshot count.\n- Rendered checks also report broken images, eager images below the fold, and\n image sources more than 2x their rendered dimensions in both axes, including\n Webstudio instance ids and measured dimensions when available.\n- Rendered checks include sanitized Resource Timing evidence and report\n browser-marked render-blocking resources plus legacy `.ttf`, `.otf`, and\n `.woff` fonts without applying a universal transfer-size threshold.\n- Fix findings through semantic mutation commands, then rerun `audit` to confirm their deterministic finding ids disappeared.\n\n## Verify dynamic bindings\n\nCommands:\n\n- webstudio verify-bindings --json\n- MCP tool: verify-bindings {"pagePath":"/pricing"}\n- MCP tool: verify-bindings {"instanceId":"<instanceId>","limit":50}\n\nNotes:\n\n- Statically checks persisted text expressions, expression/action/resource/parameter props, resource expressions, and page metadata.\n- Findings distinguish invalid syntax, unknown or out-of-scope variables, stale internal data-source ids, and missing resource or parameter references.\n- Page and instance filters can be combined when the instance belongs to the selected page. Continue findings with `cursor`.\n- This operation does not resolve rendered values or execute external resources. Preview representative loading, empty, error, and populated states after static findings are fixed.\n\n## Refactor targeted content\n\nCommands:\n\n- MCP tool: list-instances {"pagePath":"/"}\n- MCP tool: list-texts {"pagePath":"/"}\n- MCP tool: update-text {"instanceId":"<instanceId>","childIndex":0,"text":"Launch faster"}\n- MCP tool: replace-text {"find":"Old headline","replace":"New headline","match":"exact","pagePath":"/pricing","limit":20}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans"}}}\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: replace-asset {"fromAssetId":"<oldAssetId>","toAssetId":"<newAssetId>"}\n- MCP tool: replace-styles {"property":"color","fromValue":{"type":"keyword","value":"red"},"toValue":{"type":"keyword","value":"blue"}}\n- MCP tool: rewrite-css-variable-refs {"map":"variables.json contents"}\n\nNotes:\n\n- Use focused reads first, then mutate only matching instances, props, metadata, resource URLs, assets, or style references. Use `replace-text` for bounded literal text changes, `replace-prop-text` for bounded static prop text, `replace-resource-text` for fixed resource names/URLs, and `update-text` for one known child or expressions.\n\n## Optimize existing project\n\nCommands:\n\n- MCP tool: list-pages {}\n- MCP tool: list-folders {}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"title":"Pricing","meta":{"description":"Plans"}}}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: list-breakpoints {}\n- MCP tool: update-breakpoint {"breakpointId":"tablet","values":{"maxWidth":1023}}\n- MCP tool: get-styles {"instanceIds":["<instanceId>"],"includeTokens":true}\n- MCP tool: update-styles {"updates":"styles.json contents"}\n- MCP tool: attach-design-token {"designTokenId":"<tokenId>","instanceIds":"instances.json contents"}\n- MCP tool: update-project-settings {"meta":{"siteName":"Acme"}}\n\nNotes:\n\n- Use this for SEO metadata, accessibility labels, responsive behavior, token consistency, and project settings.\n\n## Connect external data\n\nCommands:\n\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"title","value":{"type":"string","value":"Hello"}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"tags","value":{"type":"json","value":["news","product"]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"filters","value":{"type":"json","value":{"tag":"news"}}}\n- MCP tool: create-resource {"resource":{"name":"Posts","method":"get","url":"https://api.example.com/posts","searchParams":[{"name":"tag","value":"filters.tag"},{"name":"source","value":{"type":"literal","value":"website"}}],"headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"posts"}\n- MCP tool: create-resource {"resource":{"name":"Post GraphQL","control":"graphql","method":"post","url":"https://api.example.com/graphql","headers":[{"name":"Content-Type","value":{"type":"literal","value":"application/json"}}],"body":"{ query: \'query Post($slug: String!) { post(slug: $slug) { title } }\', variables: { slug: system.params.slug } }"},"scopeInstanceId":"<instanceId>","dataSourceName":"post"}\n- MCP tool: create-resource {"resource":{"name":"Current Date","control":"system","method":"get","url":"/$resources/current-date","headers":[]},"scopeInstanceId":"<instanceId>","dataSourceName":"currentDate"}\n- MCP tool: update-resource {"resourceId":"<resourceId>","values":{"url":"https://api.example.com/posts"}}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.collection>{/_ collection content _/}</ws.collection>"}\n\nNotes:\n\n- Use this for CMS sections, blog listings, Ghost/headless CMS pages, n8n-style integrations, and API URLs built from variables.\n- For read data, expose GET resources as scoped data variables with `scopeInstanceId`/`dataSourceName` and read the loaded result from the resource result wrapper, usually `.data`.\n- For writes, webhooks, GraphQL submissions, and deletes, prefer unscoped resources bound to Form `action` props so they become action resources instead of auto-loaded read resources.\n- Use direct props for fixed values and prop bindings only when a prop must read a data variable, resource, action, or documented runtime context value such as `system`.\n\n## Render an array or object as repeated content\n\nCommands:\n\n- MCP tool: insert-collection {"parentInstanceId":"<instanceId>","data":{"type":"expression","value":"posts.data.items"},"itemFragment":"<ws.element ws:tag=\'article\'><ws.element ws:tag=\'h2\'>{expression`collectionItem.title`}</ws.element></ws.element>"}\n- MCP tool: inspect-instance {"instanceId":"<collectionId>","include":["props","bindings","children"]}\n\nNotes:\n\n- Use Collection whenever an array or object from a resource or data variable should render a list, grid, cards, table rows, options, tabs, or other repeated UI.\n- Pass `insert-collection` the complete array or object. Do not pass the resource response wrapper or one indexed item. External resource arrays are commonly nested under the scoped resource result\'s `data` field or deeper.\n- Pass one repeated-item Webstudio JSX root. The command creates the Collection, its private current-item/current-key parameters, the iterable binding, and descendant item bindings atomically.\n- Collection renders the item root once for every entry. Use `expression` values such as `collectionItem.title` in descendant text and props. Object iteration also exposes `collectionItemKey`.\n- Wrap multiple repeated sibling instances in one Element inside Collection.\n- For repeated Radix items such as accordion items, tabs, or menu options, bind a stable unique id or slug to every required `value` prop.\n- See the [Collection documentation](https://docs.webstudio.is/university/core-components/collection) for the equivalent Builder workflow.\n\n## Support dynamic runtime behavior\n\nCommands:\n\n- MCP tool: integrate-runtime-ui {"parentInstanceId":"<instanceId>","resources":[{"resource":{"name":"Seats","method":"get","url":"https://api.example.com/seats","headers":[]},"dataSourceName":"Seats","exposeAsDataSource":true}],"structure":{"type":"collection","data":{"type":"expression","value":"Seats.data"},"itemFragment":{"children":[{"type":"id","value":"seat"}],"instances":[{"type":"instance","id":"seat","component":"Text","children":[{"type":"expression","value":"collectionItem.label"}]}],"props":[],"dataSources":[],"resources":[],"styleSources":[],"styleSourceSelections":[],"styles":[],"breakpoints":[],"assets":[]}},"retainedBehavior":[{"instanceId":"<scriptInstanceId>","responsibility":"Seat selection behavior"}]}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n- MCP tool: create-resource {"resource":{"name":"Seats","method":"get","url":"https://api.example.com/seats","headers":[]}}\n- MCP tool: snapshot {"include":["instances","props","resources"]}\n- MCP tool: apply-patch {"baseVersion":"<version>","transactions":"patch.json contents"}\n\nNotes:\n\n- Use `integrate-runtime-ui` to create variables/resources, insert one editable fragment or Collection, and add safe data bindings in one transaction.\n- List existing script-owned responsibilities under `retainedBehavior`. The operation preserves those instances and never evaluates or accepts replacement script bodies.\n- `unsupportedConversions` records behavior that cannot be represented safely. Dry-run returns the complete transaction and the same retained/unsupported report without changing the project.\n- New actions and HtmlEmbed scripts are intentionally rejected. Create normal editable components and data bindings; keep opaque runtime behavior in existing script instances.\n\n## Build authenticated pages\n\nCommands:\n\n- MCP tool: meta.guide {"brief":"Build a Supabase-authenticated account page","taskScope":"structural-project-change","workflow":"authenticated-page"}\n- MCP tool: create-page {"name":"Account","path":"/account"}\n- MCP tool: update-page {"pageId":"<pageId>","values":{"meta":{"auth":{"method":"basic","login":"<login>","password":"<password>"}}}}\n- MCP tool: create-resource {"resource":{"name":"Session","method":"get","url":"https://api.example.com/session","headers":[]}}\n- MCP tool: create-variable {"scopeInstanceId":"<instanceId>","name":"user","value":{"type":"json","value":{}}}\n- MCP tool: update-props {"updates":"props.json contents"}\n- MCP tool: bind-props {"bindings":"bindings.json contents"}\n\nNotes:\n\n- Inspect and reuse the project\'s existing auth convention before authoring. Do\n not add a second provider or session model implicitly.\n- Model signed-out, loading, signed-in, and failed-auth states explicitly.\n- Never store credentials, service-role keys, refresh tokens, private session\n values, or authenticated response bodies in project data, command output,\n screenshots, agent instructions, or error reports. Privileged provider calls\n and authorization enforcement belong server-side.\n- Basic auth is semantic today. Provider-specific Supabase/Firebase setup still\n uses the existing resource, variable, prop, binding, and embed tools; there is\n no provider-specific installer.\n\n## Generate from design input\n\nCommands:\n\n- MCP tool: meta.guide {"brief":"Recreate this Figma design as a responsive page","taskScope":"visual-change","workflow":"design-input"}\n- MCP tool: create-page {"name":"Landing","path":"/landing"}\n- MCP tool: create-design-token {"tokens":"tokens.json contents"}\n- MCP tool: define-css-variable {"vars":"vars.json contents"}\n- MCP tool: list-breakpoints {}\n- MCP tool: insert-fragment {"parentInstanceId":"<instanceId>","fragment":"<ws.element ws:tag=\'section\'><ws.element ws:tag=\'p\'>Section copy</ws.element></ws.element>"}\n- MCP tool: update-styles {"updates":[{"instanceId":"<instanceId>","breakpoint":"<breakpointId-from-list-breakpoints>","property":"padding-left","value":{"type":"unit","unit":"px","value":24}}]}\n- MCP tool: preview.start {"host":"127.0.0.1","port":5173}\n- MCP tool: screenshot {"path":"/landing","output":"landing-desktop.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"path":"/landing","output":"landing-mobile.png","viewport":{"width":390,"height":844},"waitUntil":"load","waitForTimeout":250}\n- MCP tool: screenshot {"baseUrl":"http://127.0.0.1:5177","path":"/landing","output":"landing-desktop.png","viewport":{"width":1440,"height":900},"waitUntil":"load","waitForTimeout":250}\n\nNotes:\n\n- Use this after the agent can inspect the supplied design. There is no direct\n Figma, screenshot, Inception, or `design.md` import command.\n- Inspect and reuse existing variables, tokens, styles, components, assets, and\n page patterns before authoring. Build semantic editable structure rather than\n flattening the design into an image or absolute-positioned approximation.\n- Ask whether the user wants visual verification unless they explicitly requested\n it. If they decline, stop after focused reads and a static audit. If they opt\n in, verify one familiar viewport inside every distinct Builder breakpoint\n range, then run rendered audit and inspect the screenshots before completion.\n\n## Cross-project maintenance\n\nCommands:\n\n- webstudio mcp run .temp/projects.json\n- webstudio mcp run .temp/projects.json --dry-run\n- webstudio mcp run .temp/projects.json --approve-mutations --concurrency 2\n\nNotes:\n\n- Put shared `calls` and a `projects` array of independently linked project roots in the existing `mcp run` manifest. Project roots are relative to the manifest file.\n- Each project uses its own config, authentication, ProjectSession storage, checkpoint, and failure boundary. Confirmed successful calls are checkpointed. Reads can resume automatically; a mutation interrupted after dispatch is reported as ambiguous and is not replayed automatically, preventing silent duplicate writes.\n- Focus the manifest on bounded reads or audits first. Use per-call `dryRun`, global `--dry-run`, or explicitly approve committed mutations with `--approve-mutations` after reviewing the manifest.\n\n# Known CLI Gaps\n\n## Provider-specific authenticated pages\n\nMissing:\nCLI supports page basic auth and generic resources/props/embeds, but not guided Supabase/Firebase auth setup.\n\nCurrent fallback:\nCall `meta.guide` with `workflow:"authenticated-page"`, then create the\npage, resources, variables, props, bindings, and embeds with existing semantic\ntools.\n\nSuggested commands:\n\n- setup-auth-page\n\n## Generate from design input\n\nMissing:\nNo command imports Figma, screenshots, Inception output, or design.md and turns it into pages/tokens/layout.\n\nCurrent fallback:\nCall `meta.guide` with `workflow:"design-input"`, let the agent inspect the supplied\ndesign, then use semantic page, token, asset, fragment, style, preview,\nscreenshot, and audit tools. Use `apply-patch` only when no semantic operation\nfits.\n\nSuggested commands:\n\n- generate-from-design\n\n## Built-in cross-project maintenance\n\nMissing:\nPublic API and CLI intentionally operate on one configured project at a time; there is no built-in multi-project discovery or loop runner.\n\nCurrent fallback:\nRun the CLI from an external script that reconfigures one project/session at a time.\n\nSuggested commands:\n\n- none\n',"manual-api":'# Webstudio API CLI Manual\n\nThe API commands operate on the single project configured by:\n\n- .webstudio/config.json: projectId\n- global Webstudio config: origin and token\n\n- Pass --json to API/discovery commands that support it. Do not add --json to top-level commands unless their help/schema documents it.\n- Never pass a project id. Commands use configured project only.\n- Read ids before writing. Do not invent ids for existing records.\n- stdout is one JSON object. stderr is diagnostics.\n- Prefer MCP semantic tools for detailed project edits. Use MCP apply-patch only when no semantic tool exists.\n\n## Start\n\n{{start}}\n\n## Read First\n\n{{readFirst}}\n\n## Project Session Cache\n\n- CLI commands use one local ProjectSession snapshot for the configured project.\n- Local-capable reads use cached namespaces when compatible and fetch only missing or stale namespaces.\n- Local-capable mutations build patches from the local snapshot, then commit with the cached build version.\n- Successful mutation commits update the local snapshot only after the remote commit succeeds.\n- Server-only commands run remotely and invalidate/refetch namespaces declared by the operation catalog.\n- Use --refresh on local-capable commands to refresh required namespaces before running.\n- Successful JSON responses include compact meta.session with operationId, buildId, version, source, committed, namespaceCounts, diagnosticCount, non-empty diagnostic summaries, and optional compatibilityVersion.\n\n## Search project data\n\nRun `webstudio search-project \'{"query":"known value"}\'` to search all Builder\nnamespaces in the local project session. Narrow it with\n`{"query":"Brand","namespaces":["styleSources","styles"]}` when only specific\nnamespaces are relevant. Results contain the current value, stable match id,\nnamespace path, owning entity, resolved references, and affected pages.\n\nValues in recognized credential fields are excluded and reported in\n`excludedSensitiveValueCount`.\n\nThe command synchronizes its required Builder namespaces, then searches locally\nand returns only matches. Namespace filters limit values matched; related\nnamespaces may still supply route and reference context, and synchronization is\nunchanged. This reduces data sent to the model, not project synchronization. It\nsearches asset metadata, not asset binaries or document contents.\n\n## CLI Capability Inventory\n\nFor a short end-consumer summary of what MCP can do, see\n`manual mcp` / `webstudio man mcp`. The MCP inventory describes the same\nproject, page, element, style, data, asset, publish, domain, and visual\nverification capabilities without internal command names.\n\n### Top-Level Commands\n\n{{topLevelCapabilityIndex}}\n\n### High-Level API Commands By Area\n\n{{apiCapabilityIndex}}\n\n### MCP Tool Operations\n\nThese are MCP tools. From a shell, call them with the shortcut form `webstudio <tool> \'<json>\'` or with the explicit form `webstudio mcp single-op-call <tool> \'<json>\'`:\n\n{{mcpOnlyCommandIndex}}\n\n## Task Recipes\n\n{{taskRecipeIndex}}\n\n## Use Case Index\n\n{{useCaseIndex}}\n\n## Known CLI Gaps\n\n{{knownCliGapIndex}}\n\n## Input File Shapes\n\n{{inputFileShapeIndex}}\n\n## Raw Patch Fallback\n\napply-patch accepts either BuildPatchTransaction[] or { "transactions": BuildPatchTransaction[] }.\n\nEach transaction has:\n\n{\n"id": "patch-transaction-label",\n"payload": [\n{\n"namespace": "projectSettings",\n"patches": [\n{ "op": "replace", "path": ["meta", "siteName"], "value": "New Site" }\n]\n}\n]\n}\n\nThe transaction id is a patch label used for optimistic synchronization. It is\nnot a Builder record id. Do not invent ids for pages, instances, props,\nbreakpoints, resources, variables, folders, assets, or other project records.\n\nPatch paths are JSON-patch-like paths into Builder store data. Map-like namespaces use ids as the first path item.\n\nSupported namespaces:\n\n- pages: redirects, page records, and folders\n- projectSettings: project-wide metadata and compiler settings\n- instances: element instances and children, including text/expression children\n- props: element props, bindings, page references, resource bindings\n- styles: CSS declarations keyed by style declaration key\n- styleSources: local style sources and reusable design tokens\n- styleSourceSelections: instance-to-style-source connections\n- dataSources: data variables, parameters, and resource data sources\n- resources: data resource definitions\n- assets: project asset records handled by the existing asset patch path\n- breakpoints: responsive breakpoints\n- marketplaceProduct: marketplace metadata\n\n## Data Sources\n\n`dataSources` is the internal Builder namespace for variables. Public API, CLI,\nand MCP tools expose it through two user-facing groups:\n\n- data variables: `list-variables`, `create-variable`, `update-variable`, and\n `delete-variable`\n- data resources: `list-resources`, `create-resource`, `update-resource`, and\n `delete-resource`\n\nFor raw `snapshot`, request the public `variables` namespace rather than the\ninternal `dataSources` name. Raw patch payloads still use `dataSources` when\napplying direct changes.\n\nVariables can be scoped to an instance. Expressions under that instance can use\nthe variable by name; nested variables with the same name mask outer variables.\nVariable values support `string`, `number`, `boolean`, and `json`. Use `json`\nfor all arrays and objects, including tags, selected categories, and nested API\nfilter state.\nParameters are internal scoped runtime values provided by pages, collections,\nor components. They are not a public authoring surface: do not create, update,\nor delete parameter records. Public tools should preserve existing parameter\nrecords and may reference documented context values such as `system` in\nexpressions where they are already in scope.\n\nResource `url` accepts plain fixed URLs and paths, for example\n`https://api.example.com/posts` or `/$resources/current-date`. Dynamic URLs can\ncombine strings and variables, for example\n`"https://api.example.com/posts?tag=" + filters.tag`. Prefer `searchParams` for\nquery parameters that should be encoded separately:\n`[{ "name": "tag", "value": "filters.tag" }]`. Header values, search parameter\nvalues, and bodies are expressions for dynamic content. For fixed text, use\n`{ "type": "literal", "value": "application/json" }`; Webstudio stores the\nrequired string expression. Headers can still read variables such as\n`"Bearer " + auth.token`, and GraphQL bodies can return objects such as\n`{ query: "...", variables: { slug: system.params.slug } }`.\n\nCreate a GET resource with `scopeInstanceId` when the fetched resource result\nshould be available as a read data variable. Scoped GET resources default to\n`exposeAsDataSource: true`, are generated into the page resource `data` map,\nand may be loaded during page rendering. Use `dataSourceName` to choose the\nvariable name.\n\nFor submit/write/action resources, create the resource without\n`scopeInstanceId`, then bind a component prop such as a Form `action` to the\nresource with `bind-props` and `binding.type: "resource"`. Prop-bound resources\nare generated into the page resource `action` map instead of the read `data`\nmap. Use this shape for POST, PUT, DELETE, webhook, and other resources that\nshould run only from an explicit form/action flow, not merely because the page\nrendered.\n\nPOST, PUT, and DELETE resources default to `exposeAsDataSource: false` even\nwhen a scope is supplied. Set `exposeAsDataSource: true` only when a write-method\nresource intentionally provides render-time data, such as a read-only GraphQL\nPOST query. A scope is required, and the result includes a warning because the\nrequest may execute during page rendering. Set `exposeAsDataSource: false` on\n`update-resource` to detach an existing render-time data source.\n\nResource `method` can be `get`, `post`, `put`, or `delete`. Use GET for read\ndata. Use POST for creates, GraphQL requests, webhooks, and form submissions.\nUse PUT for full updates/replacements. Use DELETE for deletion actions.\nOptional `control` values are `graphql` and `system`: `graphql` marks a\nGraphQL-style resource, usually POST with a query body; `system` marks a\nresource intended to use the built-in `system` parameter or one of the built-in\nlocal resource URLs: `"/$resources/sitemap.xml"`,\n`"/$resources/current-date"`, and `"/$resources/assets"`. The system parameter\nfields are `system.origin`, `system.pathname`, `system.params`, and\n`system.search`.\n\nUse prop bindings for dynamic values that read variables or resources; use\ndirect props for static values.\n\nCommit raw patch:\n\nMCP tool: apply-patch\n\n## Raw Patch Examples\n\nRename the site:\n\n[\n{\n"id": "patch-site-name",\n"payload": [\n{\n"namespace": "projectSettings",\n"patches": [\n{ "op": "add", "path": ["meta", "siteName"], "value": "Acme Studio" }\n]\n}\n]\n}\n]\n\nUpdate page title metadata:\n\n[\n{\n"id": "patch-page-title",\n"payload": [\n{\n"namespace": "pages",\n"patches": [\n{ "op": "replace", "path": ["pages", "page-id", "meta", "title"], "value": "Pricing" }\n]\n}\n]\n}\n]\n\nUpdate a text child on an element:\n\n[\n{\n"id": "patch-text",\n"payload": [\n{\n"namespace": "instances",\n"patches": [\n{ "op": "replace", "path": ["instance-id", "children", 0, "value"], "value": "Launch faster" }\n]\n}\n]\n}\n]\n\nCreate records with semantic operations such as create-variable,\ncreate-resource, create-design-token, create-page, create-folder,\nand create-breakpoint. Raw patch rejects generated record\ncreation, collection replacement, record replacement with a different `id`, and\nrecord id field mutations in id-keyed namespaces because Webstudio must generate\nand preserve record ids.\n\n## Safety Rules\n\n- For MCP apply-patch, read the latest version with MCP snapshot before writing.\n- Reuse ids from MCP snapshot output when updating existing records.\n- Do not create generated records, replace generated record collections, replace records with different ids, or mutate record id fields with raw patch. Use semantic create operations so Webstudio generates ids.\n- If apply-patch reports a version conflict, read the latest build and regenerate the patch.\n- Prefer semantic MCP read tools for discovery, then use MCP snapshot for exact patch paths.\n\n## Command Index\n\n{{commandIndex}}\n',"manual-llm":@@ -12,7 +12,7 @@ export const cliDocs = {"mcp-startup-epilogue":'If you are inside the Webstudio monorepo, use the local CLI exactly as `node packages/cli/local.js ...` as your first command path. Do not use `packages/cli/bin.js` for local source-tree work; it is the packaged build entry and may use stale built output.\n\nPlain `webstudio mcp` starts the stdio MCP server for real MCP clients. Do not manually type or pipe raw JSON-RPC frames into it from an interactive shell or PTY. From a shell, use shortcut calls such as `webstudio meta.index` and `webstudio insert-fragment \'<json>\' --dry-run` for one bounded tool call. The explicit equivalent is `webstudio mcp single-op-call <tool> \'<json>\'`. Use `webstudio mcp run \'[{"tool":"components.find","input":{"brief":"button"}}]\'` for small multi-call batches in one shared CLI session. For large batches, pass a normal JSON file path such as `.temp/mcp-calls.json`. Do not use shell process substitution like `<(...)`; use inline JSON or a real file.\n\nPut each local CLI call in its own shell command; do not chain helper commands with `&&`, `;`, command substitution, or shell wrappers around the CLI when reporting a CLI step. Do not use `pnpm exec webstudio`, `pnpm --filter webstudio exec webstudio`, or a global `webstudio`; those can resolve an older binary. For example:\n\n```sh\nnode packages/cli/local.js mcp single-op-call meta.index\nnode packages/cli/local.js mcp single-op-call meta.get-more-tools \'{"tools":["insert-fragment"]}\'\nnode packages/cli/local.js mcp single-op-call insert-fragment --input-file .temp/insert-fragment.json --dry-run\n```\n\nSave the input as `.temp/insert-fragment.json`; use single quotes for JSX attributes so the JSON needs no backslash escaping:\n\n```json\n{\n "parentInstanceId": "parent-id",\n "fragment": "<ws.element ws:tag=\'section\' ws:style={css`padding: 32px; display: grid; gap: 12px;`}><ws.element ws:tag=\'h2\'>Launch Kit</ws.element><ws.element ws:tag=\'p\'>A focused section created with Webstudio JSX.</ws.element><ws.element ws:tag=\'button\'>Get started</ws.element></ws.element>"\n}\n```\n\nThe shorter local shortcut form is preferred for simple shell steps:\n\n```sh\nnode packages/cli/local.js meta.index\nnode packages/cli/local.js meta.get-more-tools \'{"tools":["insert-fragment"]}\'\nnode packages/cli/local.js insert-fragment --input-file .temp/insert-fragment.json --dry-run\n```\n\nFor simple authored/styled sections, run the three commands above. Do not grep source files, dump full MCP resources, or write parser scripts first. Use `list-pages`, `get-page-by-path`, or `list-instances` only when you still need the target `parentInstanceId`.\n\nWhen building for a Content-mode editor, use a Content Block (`ws:block`) around every region that editor must change. Content-mode text and supported props are editable only in Content Block descendants; content outside is read-only. Keep reusable insertable source options in the block\'s `ws:block-template` child. Templates themselves are protected; an editor\'s inserted copy becomes editable Content Block body content. Verify this structure before handoff.\n\nFor an MDX-backed Content Block, use the dedicated content-source tools instead of generic prop mutations: connect, switch, inspect, edit, update frontmatter, reload, or disconnect the source. Confirm before replacing existing Body content. Prefer Markdown whenever it can represent the component and all authored properties. Use `<ws.element ws:tag="tag">` for a standard HTML element with authored properties Markdown cannot express. Use `<ws.element ws:name="Template name">` only for a uniquely named top-level template. Preserve invalid source and unresolved names, inspect all diagnostics, and reload before retrying a revision conflict. A frontmatter update replaces the complete property map; preserve fields the user did not ask to remove. Only exact direct frontmatter bindings may use `mode:"readwrite"`. After a template rename or deletion, use `migrate-content-block-template-references` to preview and confirm changes across selected MDX files.\n\nRun it from the linked Webstudio project root. The startup status line prints that absolute root; use it for local files such as `<project root>/.temp/script.mjs`, screenshots, and generated artifacts.\n\nFor experiments, pass `--dry-run` to local-capable mutation calls. Read the computed transaction from `meta.session.transaction` and its base build version from `meta.session.version`. Copying a `.webstudio` folder is not an isolated project clone; `.webstudio/config.json` still points to the same remote project, so non-dry-run mutations can commit to that project.\n\nStartup marks cached ProjectSession data stale so MCP tools read the current Builder dev build.\n\nAfter startup, use focused discovery tools with concrete JSON. For delegated design-system or “use every component” tasks, start with exactly one command: `webstudio workflow.next \'{"goal":"design-system-page"}\'`; report the returned checkpoint to the parent/user and stop until continued. For other tasks, read `meta.index` first. From a shell, `webstudio mcp list-tools` is a concise alias for the initial tool catalog. Then call shortcuts such as `webstudio meta.guide \'{"brief":"Create a design system page using every component"}\'`, `webstudio workflow.next \'{"goal":"design-system-page"}\'`, `webstudio meta.get-more-tools \'{"tools":["insert-fragment"]}\'`, `webstudio components.list \'{"source":"all"}\'`, `webstudio components.summary`, `webstudio components.search \'{"brief":"radix select"}\'`, `webstudio components.get \'{"component":"@webstudio-is/sdk-components-react-radix:Select"}\'`, `webstudio templates.list`, and `webstudio templates.get \'{"component":"@webstudio-is/sdk-components-react-radix:Select"}\'`. Use `webstudio mcp single-op-call` when you need the explicit MCP form. Prefer `insert-fragment` for authored/styled sections; use `insert-component` only when inserting one automatic component template. In JSX, use `ws:style={css\\`...\\`}`for styles,`class`/`for`instead of`className`/`htmlFor`, `ActionValue`for actions instead of JavaScript functions such as`onClick={() => ...}`, only JSON-compatible plain prop values, and no host globals or dynamic code APIs. For bounded multi-step shell work, run inline JSON like `webstudio mcp run \'[{"tool":"meta.index"},{"tool":"components.search","input":{"brief":"button"}}]\'`; this reuses one CLI session without raw JSON-RPC. For larger batches, write `{ "calls": [{ "tool": "..." }] }`to`.temp/mcp-calls.json`and run`webstudio mcp run .temp/mcp-calls.json`. If any call returns `checkpoint.required`, `mcp run`stops immediately before later calls; report the checkpoint and call`checkpoint.ack {"reported":true,"continueAfterReport":true,"summary":"<what you reported>"}`before continuing. For design-system or “use every component” tasks, call`workflow.next {"goal":"design-system-page"}`for the next bounded phase, report its checkpoint, wait until the parent/user continues, call`checkpoint.ack {"reported":true,"continueAfterReport":true,"summary":"<what you reported>"}`, then run only that phase. Start coverage phases with compact`webstudio components.coverage-plan`, checkpoint, then request component coverage details with `components.coverage-plan {"detail":"roots"}`or`components.coverage-plan {"detail":"parts"}`. Do not pass `detail`to`list-pages`; use `list-pages {}`or`get-page-by-path`for page lookup. Read MCP overview resources`webstudio://project/guide`, `webstudio://project/tools-overview`, and `webstudio://project/components-overview`when useful. Read full resources`webstudio://project/tools`and`webstudio://project/components` only when focused tools are insufficient; do not dump or parse full discovery JSON for common lookup.\n\nAfter startup, MCP clients discover capabilities with `tools/list`, `resources/list`, `meta.index`, `meta.guide`, and `meta.get-more-tools`.\n\nstdout is reserved for MCP JSON-RPC messages while the server is running.\n\nThe server advertises MCP `logging` capability and sends sparse `notifications/message` logs for ready state and tool lifecycle checkpoints such as `tool preview.start started`, `tool preview.start still running after 10000ms`, and `tool preview.start succeeded in 1234ms`. It also writes the same sparse lifecycle status lines to stderr, prefixed with `[webstudio mcp]`, including a final ready line. Treat the process as healthy and long-running after either ready signal; surface tool lifecycle logs as progress checkpoints and interact through MCP JSON-RPC on stdin/stdout.\n\n`webstudio mcp single-op-call` also writes sparse lifecycle lines to stderr for each one-shot call. Use those stderr lines as progress checkpoints and parse stdout as JSON on both success and failure. Failed one-shot calls print `{ "ok": false, "error": { "code": "...", "message": "..." }, "meta": ... }` to stdout and exit nonzero. If a one-shot call returns checkpoint.required, read-only discovery remains available, while later mutations fail with `CHECKPOINT_REQUIRED` until you report the checkpoint and call `webstudio mcp single-op-call checkpoint.ack \'{"reported":true,"continueAfterReport":true,"summary":"<what you reported>"}\'`.\n\n`webstudio mcp run` also writes sparse lifecycle lines to stderr and prints one JSON result on success or failure. It stops at the first failed call or checkpoint requirement, including `CHECKPOINT_REQUIRED`, and preserves partial results so delegated agents can report exactly where they stopped.\n\nWhen `report-issue` is available, use it when the user explicitly asks to report a problem. Also use it automatically after a Webstudio-controlled tool crash, hang, malformed result, misleading success, repeated failure of the same documented operation, failed documented recovery, or required undocumented workaround. Do not automatically report expected confirmation/checkpoint flow, a clearly explained invalid input, missing permissions, `PROJECT_SESSION_BUSY`, a recovered version conflict, rate limiting, cancellation, or an offline environment. Write a complete anonymous report with the generalized user story, attempted workflow, expected behavior, actual result, recovery attempts, impact, technical context, acceptance criteria, exact model identifier, and reasoning effort. Never include names, usernames, emails, phone numbers, organizations, project or resource ids, domains, URLs, IP addresses, local paths, credentials, tokens, customer content, exact unique values, or raw contextual tool data.\n\nIf you are running as a delegated or non-streaming agent whose parent cannot see live stderr/stdout, do not batch many MCP calls silently and do not run long shell loops of shortcut or `webstudio mcp single-op-call` commands. Treat each parent-visible checkpoint as the unit of work. If the parent asks for status within 30 seconds, run exactly one shortcut command such as `webstudio meta.index` or one explicit `webstudio mcp single-op-call` command, return a concise checkpoint with that command/result, and wait for the parent to continue. Do not take a broad task such as creating a full design-system page as one execution unit. Call `workflow.next {"goal":"design-system-page"}`, report the returned phase/checkpoint, wait until the parent/user continues, call `checkpoint.ack {"reported":true,"continueAfterReport":true,"summary":"<what you reported>"}`, complete exactly that bounded phase, and return: discovery, page creation, one dry-run JSX section, one committed JSX section, one `components.coverage-insert-next` call, or one presentation pass. Phase commands do not include nextPhase in their own output. After the parent continues, acknowledge the previous checkpoint first, then call `workflow.next` with the next phase. For all-component design-system pages, checkpoint after workflow planning, discovery, page creation, call `components.coverage-insert-next` once before checkpointing again, then finish with the `presentation-pass` workflow phase.\n',"mcp-vision":- '# MCP Vision Verification\n\n## Generated App Dependency Notes\n\n- `preview.start` and `webstudio preview` install generated app dependencies under `.webstudio/preview` and reuse them across regenerations.\n- Session previews download missing project assets into `.webstudio/assets`. If `PREVIEW_ASSET_DOWNLOAD_FAILED` occurs, restore network and project asset access, then retry `preview.start`.\n- When launcher metadata is available, dependency installation reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.\n- When npm is selected, dependency installation honors `npm_config_cache`, including a caller-provided writable cache on Windows.\n- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.\n- If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.\n\n## Visual Verification Rule\n\nAsk the user whether they want visual verification before starting it, unless the user explicitly requested screenshots, visual verification, or a rendered audit. Never start `preview.start`, `screenshot`, `screenshot.diff`, `vision.install-ocr`, or a rendered audit automatically. If the user does not opt in, use focused non-visual assertions and report that vision was not run. After the user opts in, use `preview.start` and `screenshot({ path })` so vision can inspect the current MCP session and verify that generated project files are current. Iterative preview is the default: it keeps one generated-site server and browser alive, regenerates changed files, and performs an ordinary page reload without Vite HMR. Use `mode: "production"` only for release-like verification; rendered audit selects it automatically. `preview.start` is long-lived and cannot be used through one-shot `mcp single-op-call`; from a shell, use `webstudio mcp run` for preview.start/screenshot/preview.stop in one shared process, or use a long-running MCP server. When a baseline exists, use screenshot.diff to get pixel regions, OCR text changes, and diff PNG artifacts.\n\nAn authenticated project share URL is used with `webstudio init --link`; it is\nnot a generated-site preview URL. Project screenshots and rendered audits use\nthe generated local preview owned by the current CLI/MCP process. Do not pass a\nBuilder/share URL to `screenshot`, even without query parameters. Use `path`\nafter starting preview, or use `baseUrl` only for an intentional generated site\nthat is already running. Path captures verify the generated-site root marker\nand fail instead of returning a screenshot of Builder chrome.\n\n## Vision Verification Loop\n\n- Ask the user whether they want visual verification unless they explicitly requested it. Stop before calling preview, capture, comparison, OCR, or rendered-audit tools until the user opts in.\n- Make focused page/content/style changes with semantic MCP tools.\n- Call preview.start once to keep the iterative generated site running. In shell-driven workflows, run preview.start, screenshot, and preview.stop inside one `webstudio mcp run` call so they share the same preview owner.\n- Read `preview.status.stale` before relying on generated output. When present, `renderedProjectVersion` identifies the last project version materialized into the preview; a stale preview refreshes automatically on the next managed screenshot or `preview.start` call.\n- {{dependency-notes}}\n- After MCP mutations, path-based screenshots regenerate the current session in place, wait for its exact project version, and normally reload the route. The server and browser remain alive. From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running generated site without starting it. Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview; a separate one-shot `single-op-call` process does not own another process\'s preview controller.\n- For multi-page work, capture each changed page by path through the same preview server, for example screenshot({ path: "/" }), screenshot({ path: "/pricing" }), and screenshot({ path: "/about" }). The screenshot tool navigates directly to the requested route; no browser click navigation is required.\n- For responsive work, call list-breakpoints first, then capture screenshots at viewport widths based on the Builder breakpoints plus a narrow mobile and desktop width.\n- Call screenshot with { path: "/" } or the changed page path and viewport such as { width: 375, height: 812 } and { width: 1440, height: 900 }. For an existing preview in another process, call screenshot with { baseUrl: "http://127.0.0.1:5177", path: "/" }. Use waitForSelector when the page has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.\n- An explicit occupied `port` fails immediately with `PREVIEW_PORT_IN_USE`. To capture a generated site already running in another process, pass its `baseUrl` with `path`; otherwise choose another port.\n- Automatic browser discovery checks system installations, configured browser paths, and Chromium installations in the Playwright browser cache.\n- The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.\n- {{diff}} When a baseline PNG exists, call screenshot.diff with baselinePath, currentPath, and outputDir for each page/viewport pair. Add expectedText when a specific visible phrase must be present; its assertions report pass/fail plus found and missing text. Add expectedVisual to set pass/fail limits for mismatch percentage, the number of changed regions, or an overall dominant color/brightness direction.\n- {{diff}} Read screenshot.diff textAnalysis: it reports OCR status plus text that appeared, disappeared, moved, changed content, or changed font/style geometry. If OCR is unavailable, expectedText assertions fail and textAnalysis reports why; ask the user for permission to install Tesseract, then call vision.install-ocr with { "confirm": true }, or rely on visual inspection.\n- Inspect every viewport PNG and any diff artifacts with vision, then compare layout, OCR text evidence, color, spacing, imagery, and responsive framing against the user intent.\n- If the screenshot does not match, apply another focused mutation and repeat screenshot verification.\n\n## Workflow Summary With Diff\n\nAfter the user explicitly requests or approves visual verification, call preview.start once to start the iterative generated-site preview, then screenshot({ path, viewport }) after focused mutations; path screenshots regenerate changed files and reload the route in the existing server and browser. For responsive work, use list-breakpoints and capture each changed page at Builder breakpoint widths plus mobile and desktop widths. Use screenshot.diff on each baseline/current page or viewport pair when a baseline exists, then inspect pixel regions, OCR textAnalysis, and PNG/diff artifacts with vision before finishing. Use mode: "production" only for release-like verification.\n\n## Workflow Summary Without Diff\n\nAfter the user explicitly requests or approves visual verification, call preview.start once to start the iterative generated-site preview, then screenshot({ path, viewport }) after focused mutations; path screenshots regenerate changed files and reload the route in the existing server and browser. For responsive work, use list-breakpoints and capture each changed page at Builder breakpoint widths plus mobile and desktop widths. Inspect every PNG with vision before finishing. Use mode: "production" only for release-like verification.\n\nEach screenshot result includes rendered `layout` metrics when the local browser\nprovides them. `layout.horizontalOverflow: true` is deterministic evidence that\nthe rendered document exceeds the requested viewport width. Use vision for\nclipping, wrapping, hierarchy, and other judgments that layout dimensions alone\ncannot establish.\n\nPass `includeImageMetrics: true` when an ordinary screenshot needs\n`layout.images`; rendered audit enables it automatically. The array includes\neach rendered image\'s Webstudio instance id when available, loading mode,\ncompletion state, natural dimensions, rendered dimensions, and document\nposition. Rendered audits use this evidence to report broken images, eager\nloading below the fold, and sources more than 2x the rendered dimensions in\nboth axes. Oversized-source results are optimization evidence, not universal\nperformance conformance.\n\nPass `includeResourceMetrics: true` when an ordinary screenshot needs sanitized\nResource Timing evidence; rendered audit enables it automatically. Resource\nmetrics contain only the URL pathname, initiator type, transfer/body sizes,\nduration, and browser-provided render-blocking status. Origins and query strings\nare omitted. Rendered audits report explicitly blocking resources and legacy\n`.ttf`, `.otf`, or `.woff` font files without applying a universal byte-size\nbudget.\n\n## Screenshot Verification Summary\n\nAfter the user explicitly requests or approves visual verification, call preview.start once inside a long-running MCP server, then use screenshot({ path, viewport }) for fast repeated checks across multiple pages. Iterative mode is the default: after MCP mutations, path screenshots regenerate changed files and reload the requested route while keeping the server and browser alive. Use mode: "production" only for release-like verification. From one-shot shell calls or another process, use screenshot({ baseUrl, path, viewport }) to capture an already-running preview/site without generating, building, starting, or restarting preview. Use path values such as "/", "/pricing", or "/about" to capture specific generated routes. For responsive work, read list-breakpoints and capture one familiar device viewport inside each Builder breakpoint range before using vision. Screenshot waits for load by default, then fonts and two layout frames; pass waitForSelector for app readiness, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout for final settling. When a baseline exists, use screenshot.diff for changed regions, OCR textAnalysis, and diff artifacts on each baseline/current screenshot pair. Outside MCP, use `webstudio screenshot --path /pricing --output pricing.png` for one temporary generated preview capture, or keep `webstudio preview` running and pass its absolute URL to `webstudio screenshot` for repeated captures.\n\n## Screenshot Diff Evidence\n\n- Pixel evidence: total mismatch, changed regions, dominant color/luminance direction, diffPath, and contextDiffPath.\n- OCR evidence: textAnalysis.status, provider, and changes for appeared/disappeared/content_changed/moved/font_changed text. expectedText adds pass/fail assertions plus found and missing current-screen text. expectedVisual adds pass/fail quantitative checks for mismatch percentage, changed-region count, and the overall dominant color/brightness direction.\n- OCR dependency: screenshot.diff uses the system tesseract binary when available. If missing, it returns ocr_unavailable_tesseract_not_found_or_failed and still returns pixel evidence.\n- OCR install: MCP cannot prompt. Ask the user first; if they agree, call vision.install-ocr with { "confirm": true }. If automatic install is unavailable, follow the returned installUrl.\n- Final judgment: OCR and pixel diff are evidence. A vision-capable model must still inspect screenshots/diff artifacts and compare the rendered result to user intent.\n',+ '# MCP Vision Verification\n\n## Generated App Dependency Notes\n\n- `preview.start` and `webstudio preview` install generated app dependencies under `.webstudio/preview` and reuse them across regenerations.\n- Session previews download missing project assets into `.webstudio/assets`. If `PREVIEW_ASSET_DOWNLOAD_FAILED` occurs, restore network and project asset access, then retry `preview.start`.\n- When launcher metadata is available, dependency installation reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.\n- When npm is selected, dependency installation honors `npm_config_cache`. If it is unset, preview uses the writable `.webstudio/preview/.npm-cache` directory instead of the user\'s default npm cache.\n- For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.\n- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.\n- If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.\n\n## Visual Verification Rule\n\nAsk the user whether they want visual verification before starting it, unless the user explicitly requested screenshots, visual verification, or a rendered audit. Never start `preview.start`, `screenshot`, `screenshot.diff`, `vision.install-ocr`, or a rendered audit automatically. If the user does not opt in, use focused non-visual assertions and report that vision was not run. After the user opts in, use `preview.start` and `screenshot({ path })` so vision can inspect the current MCP session and verify that generated project files are current. Iterative preview is the default: it keeps one generated-site server and browser alive, regenerates changed files, and performs an ordinary page reload without Vite HMR. Use `mode: "production"` only for release-like verification; rendered audit selects it automatically. `preview.start` is long-lived and cannot be used through one-shot `mcp single-op-call`; from a shell, use `webstudio mcp run` for preview.start/screenshot/preview.stop in one shared process, or use a long-running MCP server. When a baseline exists, use screenshot.diff to get pixel regions, OCR text changes, and diff PNG artifacts.\n\nAn authenticated project share URL is used with `webstudio init --link`; it is\nnot a generated-site preview URL. Project screenshots and rendered audits use\nthe generated local preview owned by the current CLI/MCP process. Do not pass a\nBuilder/share URL to `screenshot`, even without query parameters. Use `path`\nafter starting preview, or use `baseUrl` only for an intentional generated site\nthat is already running. Path captures verify the generated-site root marker\nand fail instead of returning a screenshot of Builder chrome.\n\n## Vision Verification Loop\n\n- Ask the user whether they want visual verification unless they explicitly requested it. Stop before calling preview, capture, comparison, OCR, or rendered-audit tools until the user opts in.\n- Make focused page/content/style changes with semantic MCP tools.\n- Call preview.start once to keep the iterative generated site running. In shell-driven workflows, run preview.start, screenshot, and preview.stop inside one `webstudio mcp run` call so they share the same preview owner.\n- Read `preview.status.stale` before relying on generated output. When present, `renderedProjectVersion` identifies the last project version materialized into the preview; a stale preview refreshes automatically on the next managed screenshot or `preview.start` call.\n- {{dependency-notes}}\n- After MCP mutations, path-based screenshots regenerate the current session in place, wait for its exact project version, and normally reload the route. The server and browser remain alive. From one-shot shell calls or another process, pass `baseUrl` with `path` to capture an already-running generated site without starting it. Use preview.stop only in the same long-running MCP server or `webstudio mcp run` process that started preview; a separate one-shot `single-op-call` process does not own another process\'s preview controller.\n- For multi-page work, capture each changed page by path through the same preview server, for example screenshot({ path: "/" }), screenshot({ path: "/pricing" }), and screenshot({ path: "/about" }). The screenshot tool navigates directly to the requested route; no browser click navigation is required.\n- For responsive work, call list-breakpoints first, then capture screenshots at viewport widths based on the Builder breakpoints plus a narrow mobile and desktop width.\n- Call screenshot with { path: "/" } or the changed page path and viewport such as { width: 375, height: 812 } and { width: 1440, height: 900 }. For an existing preview in another process, call screenshot with { baseUrl: "http://127.0.0.1:5177", path: "/" }. Use waitForSelector when the page has a reliable ready marker, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout only for final visual settling.\n- An explicit occupied `port` fails immediately with `PREVIEW_PORT_IN_USE`. To capture a generated site already running in another process, pass its `baseUrl` with `path`; otherwise choose another port.\n- Automatic browser discovery checks system installations, configured browser paths, and Chromium installations in the Playwright browser cache.\n- The screenshot timeout bounds browser capture after the preview is ready. A timeout returns `SCREENSHOT_TIMEOUT`, resets the reusable browser session, and releases the shared preview lifecycle for cleanup.\n- {{diff}} When a baseline PNG exists, call screenshot.diff with baselinePath, currentPath, and outputDir for each page/viewport pair. Add expectedText when a specific visible phrase must be present; its assertions report pass/fail plus found and missing text. Add expectedVisual to set pass/fail limits for mismatch percentage, the number of changed regions, or an overall dominant color/brightness direction.\n- {{diff}} Read screenshot.diff textAnalysis: it reports OCR status plus text that appeared, disappeared, moved, changed content, or changed font/style geometry. If OCR is unavailable, expectedText assertions fail and textAnalysis reports why; ask the user for permission to install Tesseract, then call vision.install-ocr with { "confirm": true }, or rely on visual inspection.\n- Inspect every viewport PNG and any diff artifacts with vision, then compare layout, OCR text evidence, color, spacing, imagery, and responsive framing against the user intent.\n- If the screenshot does not match, apply another focused mutation and repeat screenshot verification.\n\n## Workflow Summary With Diff\n\nAfter the user explicitly requests or approves visual verification, call preview.start once to start the iterative generated-site preview, then screenshot({ path, viewport }) after focused mutations; path screenshots regenerate changed files and reload the route in the existing server and browser. For responsive work, use list-breakpoints and capture each changed page at Builder breakpoint widths plus mobile and desktop widths. Use screenshot.diff on each baseline/current page or viewport pair when a baseline exists, then inspect pixel regions, OCR textAnalysis, and PNG/diff artifacts with vision before finishing. Use mode: "production" only for release-like verification.\n\n## Workflow Summary Without Diff\n\nAfter the user explicitly requests or approves visual verification, call preview.start once to start the iterative generated-site preview, then screenshot({ path, viewport }) after focused mutations; path screenshots regenerate changed files and reload the route in the existing server and browser. For responsive work, use list-breakpoints and capture each changed page at Builder breakpoint widths plus mobile and desktop widths. Inspect every PNG with vision before finishing. Use mode: "production" only for release-like verification.\n\nEach screenshot result includes rendered `layout` metrics when the local browser\nprovides them. `layout.horizontalOverflow: true` is deterministic evidence that\nthe rendered document exceeds the requested viewport width. Use vision for\nclipping, wrapping, hierarchy, and other judgments that layout dimensions alone\ncannot establish.\n\nPass `includeImageMetrics: true` when an ordinary screenshot needs\n`layout.images`; rendered audit enables it automatically. The array includes\neach rendered image\'s Webstudio instance id when available, loading mode,\ncompletion state, natural dimensions, rendered dimensions, and document\nposition. Rendered audits use this evidence to report broken images, eager\nloading below the fold, and sources more than 2x the rendered dimensions in\nboth axes. Oversized-source results are optimization evidence, not universal\nperformance conformance.\n\nPass `includeResourceMetrics: true` when an ordinary screenshot needs sanitized\nResource Timing evidence; rendered audit enables it automatically. Resource\nmetrics contain only the URL pathname, initiator type, transfer/body sizes,\nduration, and browser-provided render-blocking status. Origins and query strings\nare omitted. Rendered audits report explicitly blocking resources and legacy\n`.ttf`, `.otf`, or `.woff` font files without applying a universal byte-size\nbudget.\n\n## Screenshot Verification Summary\n\nAfter the user explicitly requests or approves visual verification, call preview.start once inside a long-running MCP server, then use screenshot({ path, viewport }) for fast repeated checks across multiple pages. Iterative mode is the default: after MCP mutations, path screenshots regenerate changed files and reload the requested route while keeping the server and browser alive. Use mode: "production" only for release-like verification. From one-shot shell calls or another process, use screenshot({ baseUrl, path, viewport }) to capture an already-running preview/site without generating, building, starting, or restarting preview. Use path values such as "/", "/pricing", or "/about" to capture specific generated routes. For responsive work, read list-breakpoints and capture one familiar device viewport inside each Builder breakpoint range before using vision. Screenshot waits for load by default, then fonts and two layout frames; pass waitForSelector for app readiness, waitUntil:"networkidle" for network-heavy pages, and waitForTimeout for final settling. When a baseline exists, use screenshot.diff for changed regions, OCR textAnalysis, and diff artifacts on each baseline/current screenshot pair. Outside MCP, use `webstudio screenshot --path /pricing --output pricing.png` for one temporary generated preview capture, or keep `webstudio preview` running and pass its absolute URL to `webstudio screenshot` for repeated captures.\n\n## Screenshot Diff Evidence\n\n- Pixel evidence: total mismatch, changed regions, dominant color/luminance direction, diffPath, and contextDiffPath.\n- OCR evidence: textAnalysis.status, provider, and changes for appeared/disappeared/content_changed/moved/font_changed text. expectedText adds pass/fail assertions plus found and missing current-screen text. expectedVisual adds pass/fail quantitative checks for mismatch percentage, changed-region count, and the overall dominant color/brightness direction.\n- OCR dependency: screenshot.diff uses the system tesseract binary when available. If missing, it returns ocr_unavailable_tesseract_not_found_or_failed and still returns pixel evidence.\n- OCR install: MCP cannot prompt. Ask the user first; if they agree, call vision.install-ocr with { "confirm": true }. If automatic install is unavailable, follow the returned installUrl.\n- Final judgment: OCR and pixel diff are evidence. A vision-capable model must still inspect screenshots/diff artifacts and compare the rendered result to user intent.\n',} as const;export const cliDocTitles = {diff --git a/packages/cli/src/docs/api-use-cases.md b/packages/cli/src/docs/api-use-cases.mdindex 68f258fe7a..a139e25200 100644--- a/packages/cli/src/docs/api-use-cases.md+++ b/packages/cli/src/docs/api-use-cases.md@@ -161,6 +161,8 @@ Notes:- Preview installs generated app dependencies under `.webstudio/preview` and reuses them across regenerations.- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.- When launcher metadata is available, Preview reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.+- Unless `npm_config_cache` is already configured, npm uses the writable `.webstudio/preview/.npm-cache` directory.+- For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.- If dependency installation fails, check the reported package-manager path and network configuration, then reinstall or update the Webstudio CLI if the problem persists.- When a baseline exists, use screenshot.diff once per baseline/current page or viewport pair to get changed regions, OCR textAnalysis, and diff artifact paths before deciding whether the result matches. Pass expectedText for explicit pass/fail current-screen text assertions with found and missing text. Pass expectedVisual for pass/fail limits on pixel mismatch percentage, changed-region count, or the overall dominant color/brightness direction.- If screenshot.diff reports OCR unavailable and the user agrees to install it, call vision.install-ocr {"confirm":true}; otherwise continue with pixel diff and visual inspection.diff --git a/packages/cli/src/docs/mcp-vision.md b/packages/cli/src/docs/mcp-vision.mdindex 7b3dd4110d..acbe2eece2 100644--- a/packages/cli/src/docs/mcp-vision.md+++ b/packages/cli/src/docs/mcp-vision.md@@ -5,7 +5,8 @@- `preview.start` and `webstudio preview` install generated app dependencies under `.webstudio/preview` and reuse them across regenerations.- Session previews download missing project assets into `.webstudio/assets`. If `PREVIEW_ASSET_DOWNLOAD_FAILED` occurs, restore network and project asset access, then retry `preview.start`.- When launcher metadata is available, dependency installation reuses a supported npm or pnpm launcher. Without launcher metadata, it defaults to npm.-- When npm is selected, dependency installation honors `npm_config_cache`, including a caller-provided writable cache on Windows.+- When npm is selected, dependency installation honors `npm_config_cache`. If it is unset, preview uses the writable `.webstudio/preview/.npm-cache` directory instead of the user's default npm cache.+- For npm cache permission errors, unset `npm_config_cache` to use the preview-local cache, then retry. No cache deletion is required.- Do not add generated-preview dependencies to the repository root `package.json` or `pnpm-lock.yaml`.- If dependency installation fails, the error includes the selected package-manager path and sanitized diagnostics. Check that path and the network configuration, then reinstall or update the Webstudio CLI if the problem persists.diff --git a/packages/cli/src/mcp-guidance.test.ts b/packages/cli/src/mcp-guidance.test.tsindex a76528d5e5..1fd2fea39e 100644--- a/packages/cli/src/mcp-guidance.test.ts+++ b/packages/cli/src/mcp-guidance.test.ts@@ -8,16 +8,8 @@ import {} from "./mcp-guidance";test("documents generated app setup for visual verification", () => {- expect(generatedAppDependencyNotes).toEqual([- expect.stringContaining("install generated app dependencies"),- expect.stringContaining("PREVIEW_ASSET_DOWNLOAD_FAILED"),- expect.stringContaining("npm or pnpm launcher"),- expect.stringContaining("npm_config_cache"),- expect.stringContaining("Do not add generated-preview dependencies"),- expect.stringContaining("selected package-manager path"),- ]);- expect(generatedAppDependencyNotes.join("\n")).not.toContain(- "WEBSTUDIO_PREVIEW_PACKAGE_MANAGER"+ expect(getVisionVerificationLoop({ includeDiff: false })).toEqual(+ expect.arrayContaining(generatedAppDependencyNotes));expect(visualVerificationRule).toContain("generated project files are current"