Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
docs(prompts): retire the store-submission campaign context now that its last order is closed
nirholas23c752a72769
diff --git a/prompts/finish/store-submissions-00-README.md b/prompts/finish/store-submissions-00-README.mddeleted file mode 100644index 0e5ef43da..000000000--- a/prompts/finish/store-submissions-00-README.md+++ /dev/null@@ -1,109 +0,0 @@-# Store Submission Prompts — three.ws MCP--Goal: get three.ws's plugins and tools listed across **every** Claude and OpenAI marketplace + the major MCP registries:--1. **Claude Connectors Directory** (remote MCP) — prompts 01–03-2. **OpenAI ChatGPT App Directory** (Apps SDK) — prompts 04–06-3. **Differentiation layer** (the 10x moat) — prompts 07–09-4. **Claude Code plugin marketplace** + **Agent Skills pack** — prompts 10–11-5. **OpenAI GPT Store** (Custom GPT + Actions) — prompt 12-6. **MCP registries & third-party directories** (Smithery, Glama, mcp.so, PulseMCP, LobeHub) — prompt 13-7. **Shared asset kit + submission tracker** — prompt 14--Each numbered prompt was a **self-contained work order** you pasted into a fresh Claude Code chat in this repo, driving one work-stream to a verified, production-complete state.--> Retirement note (2026-07-30): every numbered prompt in this pack has been retired from the working tree and remains readable in git history. The run-order table below is kept as the map of what each work-stream owned; rows without a SHIPPED marker were retired without one, so do not read a missing marker as unfinished work. The live state of every submission target, including what still needs a human, is [`_generated/TRACKER.md`](../store-submissions/_generated/TRACKER.md), not this file. `_generated/` also holds the machine-written submission evidence and is never delete-eligible.--> **The one runnable work order in this pack is [01-submission-closeout.md](store-submissions-01-submission-closeout.md)** (added 2026-08-01). It re-verifies every tracker row against production, closes the remaining code gaps (the `@x402/mcp` `mcp://` interop bug, the Claude submission package, the devnet mint E2E), keeps the canonical metadata and directory docs consistent, and ends with one list of the steps only a human can click. Run it before any submission goes out; the tracker's numbers are only as fresh as its last run.-------## Strategic summary (read once)--The two stores are very different problems:--- **Claude Connectors Directory — achievable now.** The server already has remote streamable-HTTP transport, OAuth 2.1, per-tool `title` + annotations, manifests, a privacy policy, and a published npm package. The work is *verify + package + write the reviewer guide*. The directory explicitly permits transactional connectors, so the x402 paid model is allowed — the reviewer just needs a free end-to-end path (`forge_free`) and clean `PaymentRequired` responses.--- **OpenAI ChatGPT App Directory — needs repackaging.** OpenAI's submission policy prohibits "tokens or credits," "embedded third-party payment solutions," and "crypto speculation schemes." The current x402-per-call model and the token/pump.fun tools are disqualifying as-is. The viable route is a **separate, free, non-crypto 3D-avatar app** exposing only the generation tools (`forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`, `forge_avatar`) with an embedded GLB-viewer component — no x402, no token tooling.--Authoritative sources (re-fetch at submission time; requirements move):-- Claude: https://claude.com/docs/connectors/building/submission-- Claude remote-server guide: https://support.claude.com/en/articles/12922490-remote-mcp-server-submission-guide-- OpenAI: https://developers.openai.com/apps-sdk/app-submission-guidelines-- OpenAI submit/maintain: https://developers.openai.com/apps-sdk/deploy/submission-------## Run order--### Track A — Claude Connectors Directory (do first)-1. `01-claude-tool-annotation-audit.md` — ✅ SHIPPED (verified 2026-07-07; prompt retired — deliverables live in the repo/`_generated/`) — every tool has a correct `title` + annotation set. (Also unblocks OpenAI.)-2. `02-claude-reviewer-access-guide.md` — ✅ SHIPPED (re-audited + closed out 2026-07-08; prompt retired — deliverables live in the repo/`_generated/`) — free verification path, graceful `PaymentRequired`, reviewer setup/test-account doc. Live re-check found the free lane (`forge_free`) currently exceeding a reviewer's patience window — see `_generated/claude-tool-call-evidence.md` and `TRACKER.md` item 8 before filing.-3. `03-claude-submission-package.md` — ✅ SHIPPED (verified 2026-07-07; prompt retired — deliverables live in the repo/`_generated/`) — privacy URL, allowed-links, server metadata, the exact form-fill content + compliance acknowledgments.--### Track B — OpenAI ChatGPT App (do after Track A, or in parallel)-4. `04-openai-free-3d-endpoint.md` — a free, non-crypto, scoped 3D-generation MCP endpoint.-5. `05-openai-apps-sdk-component.md` — ✅ SHIPPED (verified 2026-07-07; prompt retired — deliverables live in the repo/`_generated/`) — the embedded GLB-viewer UI component the Apps SDK expects.-6. `06-openai-submission-package.md` — OpenAI directory metadata, policy-compliance audit, screenshots, developer verification.--### Track C — Differentiation layer (the 10x–100x; what makes us NOT just another listing)-Baseline (01–06) gets us *listed*. Track C makes us *un-ignorable*. The thesis: three.ws is the only platform in either store with generation + agent payments + agent coordination in one place — so we ship the new primitive nobody else can: **embodied, economically-autonomous agents, and 3D as a native conversational medium.**-7. `07-embodied-agent-live-avatar.md` — a persistent 3D body that lip-syncs, emotes, and idles inline. The consumer screenshot moment. (Both stores.)-8. `08-live-agent-commerce.md` — autonomous discover→reputation→pay→delegate, made *visible* with provenance receipts. The agent-economy demo. (Claude/paid only.)-9. `09-conversational-remixable-3d.md` — iterate on a model by talking to it ("bigger helmet"), plus remixable assets with provenance + automatic creator royalties. (Iteration: both stores. Royalties: Claude/paid only.)--The 100x meta-move (positioning, not a single prompt): don't just *be in* the stores — **own a category across them.** Publish an open pattern for 3D-native / spatial MCP responses (tool results that return live interactive scenes, not text) and x402 agent-commerce patterns, and be the reference implementation other apps call under the hood. Infrastructure, not an app.--### Track D — Claude plugin & skill ecosystem (beyond the Connectors Directory)-10. `10-claude-code-plugin-marketplace.md` — ✅ SHIPPED (verified 2026-07-07; prompt retired — deliverables live in the repo/`_generated/`) — finish, validate, and make installable the three.ws Claude Code plugin marketplace (`/plugin marketplace add`), incl. the missing 3D plugin.-11. `11-claude-agent-skills-pack.md` — harden the wallet/x402 skills into a distributable Agent Skills pack and add the 3D-creation skills. Tags a crypto-free creative subset the OpenAI track reuses.--### Track E — OpenAI GPT Store + MCP registries-12. `12-openai-gpt-store-actions.md` — a Custom GPT backed by an OpenAPI 3.1 Actions schema over the free, zero-crypto 3D endpoints. The fastest OpenAI listing.-13. `13-mcp-registry-directories.md` — get every three.ws MCP server listed + consistent across the official registry, Smithery, Glama, mcp.so, PulseMCP, LobeHub, from one canonical metadata source.--### Track F — Shared (run early, finish last)-14. `14-cross-store-asset-kit.md` — ✅ SHIPPED (verified 2026-07-07; prompt retired — deliverables live in the repo/`_generated/`) — one canonical icon/screenshot/copy kit (Claude-full + OpenAI-free variants) every listing reuses, plus a live submission tracker across all targets.--### Track G — Net-new primitives at the 3D × AI × crypto × web3 intersection (the build layer)-Tracks A–F get us *listed, packaged, and distributed*. Track G is where we **build tools people actually use** by fusing the four things only three.ws has in one place — generation, embeddable 3D, Solana/x402 rails, and agent identity. Each is a standalone, shippable primitive that also becomes a flagship use case for the listings above.-16. `16-tokenized-3d-nft.md` — mint a generated avatar as a real on-chain Solana NFT whose media is a live rigged viewer, with baked provenance + enforced royalties. (Claude/paid.)-17. `17-embodied-onchain-identity.md` — a persona bound to a real Solana wallet + ERC-8004 identity, whose 3D body visually encodes its chain state; tip/send within caps. (Claude/paid.)-18. `18-token-gated-3d-embeds.md` — holder-only interactive 3D embeds, gated by a server-verified on-chain balance (canonical: $THREE; runtime mint as plumbing). (Claude/paid.)-19. `19-verifiable-ai3d-provenance.md` — signed, on-chain-anchored content credentials for AI 3D; a free, coin-clean verify path that ships in BOTH stores. (Anchor: Claude/paid · Verify: both.)-20. `20-spatial-mcp-standard.md` — the open spec + reference renderer + validator for 3D-native MCP responses; three.ws as the reference implementation. (Infra; both.)-21. `21-ar-ready-exports.md` — GLB→USDZ / Scene Viewer one-tap "View in AR" links; pure consumer value, zero crypto. (Both stores.)--**Sequencing:** ship 01–06 first (real listings beat vision). Then 07 is the highest-leverage single add — it's the headline screenshot for both listings. 08 and 09 deepen the moat. Run **14 early** so 03/06/07/10/12/13 reuse its assets. Track D (10–11) and Track E (12–13) are independent of A/B and can run in parallel — 10 and 12 are the two highest-ROI additions (a one-command Claude install and the fastest OpenAI listing). 13 is mostly hygiene on the 38 servers already published. Track G (16–21) is the build pipeline: **19 + 20 + 21 are coin-clean and reusable across both stores** (ship early — they also strengthen the OpenAI listing); **16 + 17 + 18 are the web3 moat** on the Claude/paid track. Highest-leverage net-new build: **21** (consumer AR, both stores), then **19** (authenticity, both stores), then **16** (on-chain ownership — the crypto headline).-------## Rules every chat must follow (from CLAUDE.md)--- **No mocks, no fake data, no placeholders, no TODOs.** Real APIs, real endpoints, finish what you start.-- **$THREE is the only coin.** Never reference any other token anywhere. The OpenAI track must contain **zero** token/crypto surface.-- **Stage explicit paths only** (never `git add -A`/`.`) — concurrent agents share this worktree. Re-check `git status` before any commit.-- **Commit/push only when the human running the chat explicitly asks.** When they do: `git push threews main` — the only push target. Never push, pull, fetch, or merge `threeD` (the retired `nirholas/3D-Agent` mirror; its `main` has diverged with foreign history).-- **Changelog:** user-visible changes get an entry in `data/changelog.json`; run `npm run build:pages` to validate.-- **Watch the `npx vercel build` trap** — it overwrites `api/*.js` with esbuild bundles. Check `head -1` of changed `api/` files for `__defProp` before committing.--## Repo orientation-- Remote MCP endpoints: `api/_mcp/` (main), `api/_mcp3d/` (3D Studio), `api/_mcpagent/` (agent), `api/_mcpbazaar/`, `api/_mcpibm/`.-- Per-tool defs (with `title` + `annotations`): the `tools/*.js` files under each `api/_mcp*/` dir.-- Published stdio server: `mcp-server/` (`@three-ws/mcp-server`), tools in `mcp-server/src/tools/`.-- Manifests: `server.json`, `server-3d.json`, `server-agent.json`, `server-bazaar.json`, `server-ibm.json`, `server-pumpfun.json`, `mcp-server/server.json`.-- Legal: `public/legal/privacy.html`, `tos.html`, `content-policy.html`.-- Deploy: `vercel.json`. OAuth: `api/_mcp/auth.js` + `/.well-known/oauth-protected-resource`.--## Retire this file when the campaign is done (required)--This file is shared context rather than a single order, so it outlives the-prompts that cite it. Delete it in the commit that closes the LAST prompt of-this campaign, once nothing else in `prompts/finish/` references it:-- grep -rl 'store-submissions-00-README' prompts/finish/- git rm prompts/finish/store-submissions-00-README.md--While any sibling prompt of this campaign is still on disk, leave this file in-place and keep it accurate instead. The shrinking directory is the only signal-to the next agent that a campaign is closed.