Skip to content

armgit / mso

Public Imported from GitHub
0

Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.

feat(mcp): expose complete ChatGPT tool surface

Rahman16860d9f210d
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index fe858a26..add94c57 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -191,7 +191,7 @@ park a visible Approve/Deny card. The complete semantic contract is
`lib/mcp/`; access is controlled by the OAuth token scope rather than Alfa approval cards.
Capability execution is transport-neutral: `lib/capabilities/` owns shared execution policy, while `lib/mcp/`, `lib/a2a/`, and agent surfaces remain adapters. Guarded filesystem/exec/project operations are exposed through narrow `lib/host/*-api.ts` facades rather than one giant host import surface.
-The exact full transport catalog and compact ChatGPT profile are generated from source in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). Project function/MCP names remain data behind `project_capabilities`, `project_function_call`, `project_mcp_tools`, and `project_mcp_call`; they never rewrite the global tool prefix. `GET /mcp` reports the live full and ChatGPT signatures.
+The exact full transport catalog and ChatGPT model profile are generated from source in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). Project function/MCP names remain data behind `project_capabilities`, `project_function_call`, `project_mcp_tools`, and `project_mcp_call`; they never rewrite the global tool prefix. `GET /mcp` reports the live full and ChatGPT signatures.
See `docs/MCP.md` for protocol/security internals, `docs/A2A.md` for peer-agent delegation, and `docs/CHATGPT-PLUGIN.md` for the
ChatGPT-facing setup and diagrams.
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 52ed25e0..2c731e68 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -4,6 +4,12 @@
Newest first. `docs/PROGRESS.md` is the source of truth for *why* a change was made;
this is the *what*, and it is what Settings → About shows as “What's new”.
+## 2026-09-10
+
+**Added**
+
+- `mcp` expose complete ChatGPT tool surface
+
## 2026-09-09
**Added**
diff --git a/docs/CHATGPT-PLUGIN.md b/docs/CHATGPT-PLUGIN.md
index 43458af4..7b04bcd9 100644
--- a/docs/CHATGPT-PLUGIN.md
+++ b/docs/CHATGPT-PLUGIN.md
@@ -8,9 +8,9 @@ The exact current server/toolset identity plus the full MSO and compact ChatGPT
Browser-hosted MCP probes from ChatGPT are allowed only when they target the configured public `OS_PUBLIC_ORIGIN`; arbitrary origins and ChatGPT-origin requests aimed at loopback remain denied. Additional trusted browser hosts can be added explicitly with comma-separated `OS_MCP_BROWSER_ORIGINS`. Public OAuth discovery and DCR responses carry CORS metadata; authenticated `/mcp` echoes only an approved exact origin.
-The compact descriptor regression currently measures **67,496 JSON bytes** for the current generated ChatGPT profile (roughly 16.9k tokens at a 4-byte/token estimate), with the largest individual descriptor **2,615 bytes**. CI keeps the profile below 72 KiB and each descriptor below 8 KiB. Bytes are the deterministic contract; token estimates vary by tokenizer.
+The full generic descriptor regression currently measures about **95 KiB JSON** for the generated ChatGPT model profile (roughly 23.5k tokens at a 4-byte/token estimate), while each individual descriptor remains below 8 KiB. CI keeps the complete profile below 96 KiB and each descriptor below 8 KiB. Bytes are the deterministic contract; token estimates vary by tokenizer.
-## Why ChatGPT gets a compact profile
+## Why ChatGPT gets the full generic profile
The full MSO catalog remains available to generic MCP clients. ChatGPT action scanning is different: it freezes names, titles, descriptions, JSON Schemas, safety annotations and security metadata into a cached action snapshot. Sending every internal MSO capability wastes scan/model context and makes refreshes more fragile.
@@ -50,7 +50,7 @@ The ChatGPT projection compacts verbose descriptions/schema descriptions without
## Exact ChatGPT model tool profile
-The exact current ChatGPT profile is generated under **ChatGPT static profile** in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). Its source is `CHATGPT_TOOL_NAMES` in `lib/mcp/tool-contract.ts`; CI fails if that profile references a missing global tool. The profile remains fail-closed at both `tools/list` and `tools/call`, while OAuth `read < write < exec` is enforced independently.
+The exact current ChatGPT profile is generated under **ChatGPT model profile** in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). It automatically projects every MSO-owned generic model/operator tool registered in `lib/mcp/tools.ts`; only the two compatibility bridges marked app-only remain outside model tool calling. Project-owned MCP names still never enter the global catalog. OAuth `read < write < exec` remains enforced independently.
The restored Original MSO operator primitives (`sys_*`, `fs_usage` + full bounded filesystem CRUD, `apps_*`, browser power/status, and Dokploy/Cloudflare/Hostinger operations) are again first-class ChatGPT actions. Fresh 3 adds `vps_status`, project snapshot/diff/history/knowledge, private project-agent message/status, connection inventory, and Convex database seams without removing the lower-level primitives. This is intentional: aggregate tools optimize common turns; bounded primitives preserve direct operator control.
@@ -104,7 +104,7 @@ Project `.mso/functions.json` remains a separate generic seam: `project_capabili
Each ChatGPT conversation is bound to a separate durable MSO AgentSession using a privacy-safe hash of the host conversation id. Raw ChatGPT conversation ids are not persisted.
-The compact profile includes:
+The ChatGPT model profile includes:
- `local_agents_list`
- `local_agent_message_send`
@@ -114,7 +114,7 @@ The compact profile includes:
`local_agent_inbox(wait_ms=1..20000)` keeps only the current foreground MCP request open, registers the existing in-process Local Agent receiver, and returns early when another same-principal session sends a message. Durable file-backed mailbox state remains authoritative and closes the read→subscribe race. There is no DB, webhook, broker, WebSocket or spawned worker in this path.
-A completely idle ChatGPT conversation still cannot be awakened by a remote MCP server. Its durable mail is delivered on its next MCP call. `local_agent_request` and `agent_subagent_run` remain full-catalog worker primitives and are intentionally not part of the compact ChatGPT profile.
+A completely idle ChatGPT conversation still cannot be awakened by a remote MCP server. Its durable mail is delivered on its next MCP call. `local_agent_request` and `agent_subagent_run` are part of the ChatGPT model profile; their existing scope, isolation, and server-side authorization rules remain authoritative.
## OAuth contract
diff --git a/docs/MCP.md b/docs/MCP.md
index 61cfce00..e4291eb8 100644
--- a/docs/MCP.md
+++ b/docs/MCP.md
@@ -114,7 +114,7 @@ restarting a daemon does not require handing one over either.
`tools/list` is filtered by the token's scope, and `tools/call` re-checks it — a
client that calls a tool it was never shown still gets refused.
-**Scope remains the permission boundary, while a client profile may advertise a smaller static catalog for compatibility/context budget.** The ChatGPT profile is fail-closed at both list and call time; it never grants anything the token scope does not already allow. Project-specific capabilities remain dynamic data rather than global names. `lib/host` path/command guards remain authoritative for every operation. Every operational tool carries an OPTIONAL
+**Scope remains the permission boundary.** The ChatGPT profile advertises the complete MSO-owned generic model/operator catalog and never grants anything the token scope does not already allow. App-only compatibility bridges remain app-only, and project-specific capabilities remain dynamic data rather than global names. `lib/host` path/command guards remain authoritative for every operation. Every operational tool carries an OPTIONAL
`workflow_id`: it correlates steps, it never gates a capability. `lib/mcp/global-tools.test.ts`
pins all of this, including that a token's visible list matches its callable set exactly.
@@ -154,7 +154,7 @@ The catalog has a stable server version plus a schema-derived toolset signature.
Settings → MCP shows the current version/hash/count and stores a browser-local acknowledgement when the operator marks ChatGPT refreshed. A later signature change becomes an explicit stale-snapshot warning. This does not mutate ChatGPT remotely; it makes the required refresh visible instead of relying on memory.
-The exact current server/toolset identity, full transport/model counts, compact ChatGPT profile, scope counts and tool names are generated in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). `GET /mcp` remains the live deployed authority and additionally exposes schema-derived full/profile hashes. See [`CHATGPT-PLUGIN.md`](./CHATGPT-PLUGIN.md) for ChatGPT-specific behavior.
+The exact current server/toolset identity, full transport/model counts, ChatGPT model profile, scope counts and tool names are generated in [`generated/MCP-CATALOG.md`](./generated/MCP-CATALOG.md). `GET /mcp` remains the live deployed authority and additionally exposes schema-derived full/profile hashes. See [`CHATGPT-PLUGIN.md`](./CHATGPT-PLUGIN.md) for ChatGPT-specific behavior.
`agent_memory_search` is the typed-memory retrieval surface. It resolves semantic/episodic/procedural claims at an optional point in time, returns confidence/provenance and competing effective claims, and can expose superseded/retracted history when explicitly requested. `agent_memory_remember` remains the write surface and now accepts typed metadata; raw ChatGPT conversation ids are never stored as provenance.
@@ -166,7 +166,7 @@ The ChatGPT client profile additionally supplies a DRY output contract for every
`toolDescriptor()` is the SSOT normalization layer for every MCP host. It guarantees a human-readable `title`, complete `readOnlyHint` / `destructiveHint` / `openWorldHint` booleans, optional `idempotentHint`, and matching top-level plus `_meta.securitySchemes`. Individual tool declarations may override safety semantics; otherwise conservative MSO defaults are applied from scope/operation class.
-The OAuth scope remains the permission boundary. A client profile may expose fewer MSO-owned generic names for compatibility/context budget, and the dispatcher rejects hidden names when called directly. ChatGPT is the first compact profile: it is derived from `CHATGPT_TOOL_NAMES`, independently hashed, and regression-tested for total/per-tool descriptor size. Full MCP clients keep the complete generic catalog.
+The OAuth scope remains the permission boundary. ChatGPT now receives the complete MSO-owned generic model/operator catalog; the dispatcher still rejects app-only compatibility bridges and unknown/project-owned global names. The ChatGPT descriptor projection is independently hashed and regression-tested for total/per-tool size. Project-owned MCP capabilities remain dynamic behind the generic project bridge.
### RASMIC repo-local orchestration memory
diff --git a/docs/generated/MCP-CATALOG.md b/docs/generated/MCP-CATALOG.md
index 6ee95b24..c1ac2be7 100644
--- a/docs/generated/MCP-CATALOG.md
+++ b/docs/generated/MCP-CATALOG.md
@@ -1,16 +1,16 @@
# Generated MCP catalog
-> **Generated file — do not edit manually.** Source of truth: `lib/mcp/tools.ts`, its registered `tools-*` modules, `lib/mcp/toolset.ts`, and the compact ChatGPT name set in `lib/mcp/tool-contract.ts`. Regenerate with `node scripts/gen-mcp-catalog.mjs`. The live deployed authority remains `GET /mcp`.
+> **Generated file — do not edit manually.** Source of truth: `lib/mcp/tools.ts`, its registered `tools-*` modules, `lib/mcp/toolset.ts`, and the ChatGPT app-only exclusions in `lib/mcp/tool-contract.ts`. Regenerate with `node scripts/gen-mcp-catalog.mjs`. The live deployed authority remains `GET /mcp`.
## Full MSO catalog
-<!-- mcp-toolset: server=1.12.3 version=2026.09.09.1 tools=97 read=47 write=30 exec=20 -->
+<!-- mcp-toolset: server=1.12.3 version=2026.09.10.1 tools=97 read=47 write=30 exec=20 -->
| Fact | Current source value |
|---|---:|
| MCP server | `1.12.3` |
-| Toolset | `2026.09.09.1` |
-| Toolset changed at | `2026-09-09T05:55:00Z` |
+| Toolset | `2026.09.10.1` |
+| Toolset changed at | `2026-09-10T02:40:00Z` |
| Transport tools | **99** |
| Model/operator tools | **97** |
| Read | **47** |
@@ -129,29 +129,37 @@
- `render_mso_surface`
- `workflow_status`
-## ChatGPT static profile
+## ChatGPT model profile
-<!-- mcp-chatgpt-profile: server=1.12.3 version=2026.09.09.1 tools=71 read=37 write=21 exec=13 app-only=2 total=73 -->
+<!-- mcp-chatgpt-profile: server=1.12.3 version=2026.09.10.1 tools=97 read=47 write=30 exec=20 app-only=2 total=99 -->
-The ChatGPT profile is a fail-closed static projection defined by `CHATGPT_TOOL_NAMES`. OAuth scope is still enforced independently; project-owned MCP tool names remain dynamic data behind the generic project bridge.
+The ChatGPT profile automatically projects the complete MSO-owned generic model/operator catalog. OAuth scope is still enforced independently; app-only compatibility bridges stay app-only, and project-owned MCP tool names remain dynamic data behind the generic project bridge.
| Fact | Current source value |
|---|---:|
-| ChatGPT transport tools | **73** |
-| ChatGPT model/operator tools | **71** |
-| Read | **37** |
-| Write | **21** |
-| Exec | **13** |
+| ChatGPT transport tools | **99** |
+| ChatGPT model/operator tools | **97** |
+| Read | **47** |
+| Write | **30** |
+| Exec | **20** |
| App-only bridges | **2** |
-### ChatGPT read (37)
+### ChatGPT read (47)
+- `a2a_agent_discover`
+- `a2a_agents_list`
+- `a2a_task_get`
+- `agent_memory_read`
+- `agent_memory_search`
- `agent_session_current`
+- `agent_session_resume`
+- `agent_sessions_list`
- `apps_list`
- `apps_logs`
- `browser_status`
- `cloudflare_zones_list`
- `connections_list`
+- `dokploy_applications_list`
- `dokploy_projects_list`
- `exec_job_status`
- `fs_list`
@@ -171,6 +179,7 @@ The ChatGPT profile is a fail-closed static projection defined by `CHATGPT_TOOL_
- `project_diff`
- `project_get`
- `project_knowledge_get`
+- `project_memory_search`
- `projects_list`
- `read_pipeline`
- `render_mso_block`
@@ -182,13 +191,20 @@ The ChatGPT profile is a fail-closed static projection defined by `CHATGPT_TOOL_
- `skills_search`
- `sys_processes`
- `sys_stats`
+- `tool_forge_candidates`
- `vps_status`
-### ChatGPT write (21)
+### ChatGPT write (30)
+- `a2a_agent_register`
+- `a2a_agent_remove`
+- `agent_memory_forget`
+- `agent_memory_remember`
+- `agent_session_note`
- `agent_session_rename`
- `apps_power`
- `cloudflare_dns_upsert`
+- `dokploy_application_public_env_upsert`
- `dokploy_project_ensure`
- `fs_copy`
- `fs_delete`
@@ -202,19 +218,27 @@ The ChatGPT profile is a fail-closed static projection defined by `CHATGPT_TOOL_
- `local_agent_message_send`
- `local_agent_reply`
- `project_knowledge_set`
+- `project_memory_upsert`
+- `project_script_run`
- `session_artifact_register`
- `session_artifacts_cleanup`
+- `tool_forge_propose`
- `workflow_cancel`
- `workflow_finish`
- `workflow_start`
-### ChatGPT exec (13)
+### ChatGPT exec (20)
+- `a2a_handoff`
+- `a2a_message_send`
+- `a2a_task_cancel`
+- `agent_subagent_run`
- `browser_power`
- `exec_job_cancel`
- `exec_job_start`
- `exec_run`
- `integration_execute`
+- `local_agent_request`
- `project_agent_run`
- `project_database_call`
- `project_database_query`
@@ -223,6 +247,8 @@ The ChatGPT profile is a fail-closed static projection defined by `CHATGPT_TOOL_
- `project_function_call`
- `project_mcp_call`
- `project_mcp_tools`
+- `tool_forge_evaluate`
+- `tool_forge_promote`
### ChatGPT app-only bridges (2)
diff --git a/docs/integrations/CHATGPT-UI.md b/docs/integrations/CHATGPT-UI.md
index e1e3893e..98eb8021 100644
--- a/docs/integrations/CHATGPT-UI.md
+++ b/docs/integrations/CHATGPT-UI.md
@@ -75,7 +75,7 @@ The canonical `resources/list` response advertises only Block and Page. Previous
4. Sandboxed Pages use the standard `tools/call` bridge for native data refreshes. ChatGPT-only helpers such as display mode, private widget state, follow-up messages, and `openExternal` are feature-detected.
5. The dedicated UI origin is deployment-derived: explicit `OS_MCP_UI_ORIGIN` wins; otherwise an `mso.<domain>` public origin derives `mso-ui.<domain>`, and other safe public origins are reused. Standard CSP fields live only in `_meta.ui.csp`; the legacy `openai/widgetCSP` object retains only `redirect_domains` for `Open in MSO`.
-Current ChatGPT transport profile: **66 tools** — 64 model-visible actions (35 read / 17 write / 12 exec) plus app-only `workflow_status` and `render_mso_surface`. The full transport catalog has 90 tools. Exactly two tools bind UI resources: `render_mso_block` and `render_mso_page`.
+Current ChatGPT model profile is source-generated from the complete MSO-owned generic model/operator catalog; exact counts live in [`../generated/MCP-CATALOG.md`](../generated/MCP-CATALOG.md). `workflow_status` and `render_mso_surface` remain app-only compatibility bridges. Exactly two model tools bind UI resources: `render_mso_block` and `render_mso_page`.
## Source boundaries
diff --git a/docs/reference/WORKSPACE-GUIDE.md b/docs/reference/WORKSPACE-GUIDE.md
index c93c2f36..8feee9af 100644
--- a/docs/reference/WORKSPACE-GUIDE.md
+++ b/docs/reference/WORKSPACE-GUIDE.md
@@ -66,7 +66,7 @@ CLI, and optional MCP interface use the same guarded host capabilities.
- **Use Alfa Cockpit** — the in-browser assistant now surfaces selected high-value MSO runtime context at the point of execution: active provider/model, searchable project context, trusted host Skills, typed-memory status, native agent sessions, same-principal local-agent presence, pending approvals, and one Activity & Runs timeline. The Cockpit intentionally does **not** claim full CLI parity; credentials/server wiring stay in Settings and dangerous actions still use Alfa's existing approval cards. The same read-model is scriptable as `mso cockpit [project]` / `mso cockpit search <query>`, so the UI does not own a parallel project/session/memory implementation.
- **Run saved Automations for real** — browser-local automation recipes now re-enter Alfa's normal tool loop instead of only narrating their steps. Reads execute through the existing host bindings; write/exec steps keep the same per-call human approval gate. Playbook starter prompts also appear as bounded quick actions in an empty chat.
- **Automate deployment providers without handing tokens to the model** — Dokploy and Cloudflare are built-in feature apps; `mso provider` also supports Hostinger DNS. Secrets stay in owner-only MSO state while bounded provider tools perform live checks and approved deployment/DNS operations. See [docs/INFRASTRUCTURE-PROVIDERS.md](../INFRASTRUCTURE-PROVIDERS.md).
-- **Drive the box from ChatGPT, Codex, Claude Code, Cursor, Gemini CLI or VS Code** — MSO exposes one generic OAuth MCP server. ChatGPT gets a source-generated compact static profile with complete title/safety/security metadata, resource-bound OAuth + rotating refresh tokens, while other MCP clients can use the full catalog. Exact current counts and hashes live in [the generated MCP catalog](../generated/MCP-CATALOG.md) so this README does not freeze a stale scanner count. Project-owned MCP tool names never enter the global list: discover/call them dynamically through the generic `project_mcp_tools` / `project_mcp_call` seam. See the [ChatGPT custom MCP guide](../CHATGPT-PLUGIN.md) and [MCP reference](../MCP.md).
+- **Drive the box from ChatGPT, Codex, Claude Code, Cursor, Gemini CLI or VS Code** — MSO exposes one generic OAuth MCP server. ChatGPT gets the complete source-generated MSO-owned generic model/operator profile with complete title/safety/security metadata and resource-bound OAuth + rotating refresh tokens; app-only bridges stay app-only and project-owned MCP names stay dynamic. Exact current counts and hashes live in [the generated MCP catalog](../generated/MCP-CATALOG.md) so this README does not freeze a stale scanner count. Project-owned MCP tool names never enter the global list: discover/call them dynamically through the generic `project_mcp_tools` / `project_mcp_call` seam. See the [ChatGPT custom MCP guide](../CHATGPT-PLUGIN.md) and [MCP reference](../MCP.md).
- **Message live local sessions by a short name** — every durable same-principal MSO session gets a unique familiar handle such as `@milo`, `@luna`, or `@nara`, independent from its longer auto-generated session title. `/rename <name>` changes that handle, while `/title <text>` changes the session description. `@milo …` resolves only agents with a current lease **and** a live receiver subscription, then creates an explicitly correlated request; replies relay deterministically into the originating conversation while notify-only events remain passive. Presence lease (`idle`/`busy`/`offline`) and receiver subscription (`consumerConnected`) are reported separately so an idle-but-unsubscribed target is diagnosable. For ChatGPT/MCP sessions, `local_agent_inbox(wait_ms=...)` temporarily turns the foreground MCP call into that live receiver for up to 20 seconds, enabling direct two-way session messaging without a worker, DB, webhook, or broker. See [Local Agents](../LOCAL-AGENTS.md).
- **Spawn focused same-session subagents** — `/spawn` and `agent_subagent_run` create bounded foreground workers with isolated context, explicit delegated scope, turn/timeout caps, no recursive spawn, and final-result-only return. They are not Local Agent peers and do not run autonomously after the parent call returns. See [Subagents](../SUBAGENTS.md).
- **Delegate between remote agents with standard A2A v1** — register public HTTPS peers, attach private API-key/Bearer/OAuth access-token profiles, stream remote tasks over SSE, or expose MSO's authenticated inbound Agent Card when a public HTTPS origin is configured. Inbound credentials are owner-minted `read | write | exec` capabilities and run in a memory-isolated task context. See [A2A](../A2A.md).
diff --git a/lib/mcp/chatgpt-profile.test.ts b/lib/mcp/chatgpt-profile.test.ts
index d19e005b..ebd97278 100644
--- a/lib/mcp/chatgpt-profile.test.ts
+++ b/lib/mcp/chatgpt-profile.test.ts
@@ -2,16 +2,18 @@ import { describe, expect, it, vi } from "vitest";
vi.mock("server-only", () => ({}));
const { dispatch } = await import("./dispatch");
-const { CHATGPT_TOOL_NAMES } = await import("./tool-contract");
+const { CHATGPT_APP_ONLY_TOOL_NAMES } = await import("./tool-contract");
+const { TOOLS } = await import("./tools");
const context = { principal: "mcp-client:chatgpt-test", sessionId: "session-profile", toolProfile: "chatgpt" as const };
-describe("ChatGPT compact MCP profile", () => {
- it("advertises only MSO-owned generic tools with complete OpenAI metadata and scanner headroom", async () => {
+describe("ChatGPT full generic MCP profile", () => {
+ it("advertises every MSO-owned generic model tool with complete OpenAI metadata and bounded scanner size", async () => {
const response = await dispatch({ id: 1, method: "tools/list" }, "exec", "test", context);
const tools = (response.result as { tools: Array<Record<string, unknown>> }).tools;
- expect(tools.map((tool) => tool.name).sort()).toEqual([...CHATGPT_TOOL_NAMES].sort());
- expect(tools).toHaveLength(CHATGPT_TOOL_NAMES.size);
+ const expected = TOOLS.map((tool) => tool.name).sort();
+ expect(tools.map((tool) => tool.name).sort()).toEqual(expected);
+ expect(tools).toHaveLength(expected.length);
for (const tool of tools) {
expect(typeof tool.title, String(tool.name)).toBe("string");
@@ -30,10 +32,17 @@ describe("ChatGPT compact MCP profile", () => {
const bytes = Buffer.byteLength(JSON.stringify(tools));
if (process.env.MSO_PROFILE_METRICS === "1") console.info(`CHATGPT_PROFILE_METRICS tools=${tools.length} bytes=${bytes} roughTokens4=${Math.ceil(bytes/4)} maxToolBytes=${Math.max(...tools.map((tool)=>Buffer.byteLength(JSON.stringify(tool))))}`);
- expect(bytes).toBeLessThan(72 * 1024);
+ expect(bytes).toBeLessThan(96 * 1024);
expect(tools.some((tool) => tool.name === "local_agent_inbox")).toBe(true);
expect(tools.some((tool) => tool.name === "project_mcp_tools")).toBe(true);
expect(tools.some((tool) => tool.name === "project_mcp_call")).toBe(true);
+ expect(tools.some((tool) => tool.name === "tool_forge_promote")).toBe(true);
+ expect(tools.some((tool) => tool.name === "a2a_handoff")).toBe(true);
+ expect(tools.some((tool) => tool.name === "project_memory_upsert")).toBe(true);
+ for (const name of CHATGPT_APP_ONLY_TOOL_NAMES) {
+ const tool = tools.find((entry) => entry.name === name) as { _meta?: { ui?: { visibility?: string[] } } } | undefined;
+ expect(tool?._meta?.ui?.visibility, name).toEqual(["app"]);
+ }
const pipeline = tools.find((tool) => tool.name === "read_pipeline") as { inputSchema?: { properties?: { calls?: { items?: { properties?: { transform?: { properties?: { where?: { items?: { properties?: { value?: unknown } } } } } } } } } } } | undefined;
expect(pipeline?.inputSchema?.properties?.calls?.items?.properties?.transform?.properties?.where?.items?.properties?.value).toEqual({
anyOf: [{ type: "string", maxLength: 2048 }, { type: "number" }, { type: "boolean" }, { type: "null" }],
@@ -41,8 +50,10 @@ describe("ChatGPT compact MCP profile", () => {
expect(tools.some((tool) => String(tool.name) === "private_echo" || String(tool.name).startsWith("project_fixture_"))).toBe(false);
});
- it("fails closed when a compact-profile client guesses a full-catalog tool", async () => {
- const result = await dispatch({ id: 2, method: "tools/call", params: { name: "tool_forge_promote", arguments: { id: "candidate-hidden" } } }, "exec", "test", context);
- expect(result.error).toMatchObject({ code: -32602 });
+ it("keeps unknown/project-owned global names out while app-only bridges remain callable by the app", async () => {
+ for (const name of ["private_echo", "project_fixture_secret_tool"]) {
+ const result = await dispatch({ id: 2, method: "tools/call", params: { name, arguments: {} } }, "exec", "test", context);
+ expect(result.error, name).toMatchObject({ code: -32602 });
+ }
});
});
diff --git a/lib/mcp/tool-contract.ts b/lib/mcp/tool-contract.ts
index efa110fe..5840d0fe 100644
--- a/lib/mcp/tool-contract.ts
+++ b/lib/mcp/tool-contract.ts
@@ -30,34 +30,13 @@ export function outputSchemaForProfile(tool: McpTool, profile: McpToolProfile =
return tool.outputSchema ?? (profile === "chatgpt" ? CHATGPT_RESULT_OUTPUT_SCHEMA : undefined);
}
-// ChatGPT scans a frozen descriptor snapshot. Keep that public snapshot deliberately
-// small; every name here is an MSO-owned generic primitive. Project-owned MCP tool
-// names are discovered/called dynamically through project_mcp_* and NEVER appended
-// to this list or TOOLS.
-export const CHATGPT_TOOL_NAMES = new Set([
- // Workflow/session intelligence. workflow_status and render_mso_surface stay app-only at presentation time.
- "workflow_start", "workflow_status", "workflow_finish", "workflow_cancel",
- "skills_search", "skills_list", "skills_read", "read_pipeline",
- "agent_session_current", "agent_session_rename", "session_artifacts", "session_artifact_register", "session_artifacts_cleanup",
- "local_agents_list", "local_agent_inbox", "local_agent_message_send", "local_agent_reply", "local_agent_request_wait",
-
- // Project-first operator surface. Project-owned names remain dynamic.
- "projects_list", "project_get", "project_changes_list", "project_diff", "project_capabilities",
- "project_knowledge_get", "project_knowledge_set", "connections_list", "project_agent_run", "project_agent_status",
- "project_mcp_tools", "project_mcp_call", "project_function_call",
- "project_database_status", "project_database_tools", "project_database_call", "project_database_query",
-
- // Original MSO bounded VPS/file/application/browser power restored to ChatGPT.
- "vps_status", "mso_surface_apps_list", "render_mso_block", "render_mso_page", "render_mso_surface", "screen_capture", "fs_list", "fs_read", "fs_search", "fs_usage",
- "fs_write", "fs_upload_file", "fs_mkdir", "fs_move", "fs_copy", "fs_delete",
- "sys_stats", "sys_processes", "apps_list", "apps_logs", "apps_power", "browser_status", "browser_power",
-
- // Infrastructure operations remain explicit and bounded; secrets stay server-side.
- "integration_query", "integration_manage", "integration_execute", "integration_setup_open", "infra_providers_list", "infra_provider_doctor", "dokploy_projects_list", "dokploy_project_ensure",
- "cloudflare_zones_list", "cloudflare_dns_upsert", "hostinger_dns_upsert",
-
- // Arbitrary shell is still a last-resort escape hatch and long builds remain job-bound.
- "exec_run", "exec_job_start", "exec_job_status", "exec_job_cancel",
+// ChatGPT should expose the complete MSO-owned generic model/operator catalog.
+// Project-owned MCP tool names are never appended to TOOLS; they remain dynamic
+// behind project_mcp_tools/project_mcp_call. These two compatibility bridges are
+// UI-only and therefore intentionally excluded from model tool calling.
+export const CHATGPT_APP_ONLY_TOOL_NAMES = new Set([
+ "workflow_status",
+ "render_mso_surface",
] as const);
const TITLES: Record<string, string> = {
@@ -111,8 +90,8 @@ export function toolSecuritySchemes(tool: McpTool): McpSecurityScheme[] {
return tool.securitySchemes ?? [{ type: "oauth2", scopes: [tool.scope] }];
}
-export function toolAllowedForProfile(name: string, profile: McpToolProfile = "full"): boolean {
- return profile === "full" || CHATGPT_TOOL_NAMES.has(name as never);
+export function toolAllowedForProfile(_name: string, _profile: McpToolProfile = "full"): boolean {
+ return true;
}
function compactText(value: string, max = 125): string {
diff --git a/lib/mcp/toolset.ts b/lib/mcp/toolset.ts
index 15464cf2..33736f6a 100644
--- a/lib/mcp/toolset.ts
+++ b/lib/mcp/toolset.ts
@@ -4,8 +4,8 @@ import type { McpTool } from "./tool-kit";
import { toolDescriptor, type McpToolProfile } from "./tool-contract";
export const MCP_SERVER_VERSION = "1.12.3";
-export const MCP_TOOLSET_VERSION = "2026.09.09.1";
-export const MCP_TOOLSET_CHANGED_AT = "2026-09-09T05:55:00Z";
+export const MCP_TOOLSET_VERSION = "2026.09.10.1";
+export const MCP_TOOLSET_CHANGED_AT = "2026-09-10T02:40:00Z";
export type McpToolsetInfo = {
serverVersion: string;
diff --git a/scripts/check-docs.mjs b/scripts/check-docs.mjs
index 683d9dbb..d38e5c27 100755
--- a/scripts/check-docs.mjs
+++ b/scripts/check-docs.mjs
@@ -9,11 +9,12 @@ const ROOT = resolve(dirname(fileURLToPath(import.meta.url)), "..");
const fail = [];
const read = (p) => readFileSync(join(ROOT, p), "utf8");
-// --- MCP catalog: one generated SSOT for the full server + compact ChatGPT profile.
+// --- MCP catalog: one generated SSOT for the full server + ChatGPT model profile.
const catalog = collectMcpCatalog(ROOT);
for (const name of catalog.orphanModules) fail.push(`lib/mcp/${name}: tool module exists but is not reachable from lib/mcp/tools.ts`);
for (const name of catalog.missingModules) fail.push(`lib/mcp/${name}: imported tool module is missing`);
for (const name of catalog.profileMissing) fail.push(`lib/mcp/tool-contract.ts: ChatGPT profile references missing MCP tool ${name}`);
+for (const name of catalog.appOnlyMissing) fail.push(`lib/mcp/tool-contract.ts: ChatGPT app-only profile references missing MCP tool ${name}`);
const appUiDoc = read("docs/integrations/CHATGPT-UI.md");
for (const name of catalog.appOnly) {
diff --git a/scripts/gen-mcp-catalog.mjs b/scripts/gen-mcp-catalog.mjs
index c83ec29a..8ea2c655 100755
--- a/scripts/gen-mcp-catalog.mjs
+++ b/scripts/gen-mcp-catalog.mjs
@@ -11,6 +11,7 @@ const errors = [
...catalog.orphanModules.map((name) => `orphan MCP tool module: lib/mcp/${name}`),
...catalog.missingModules.map((name) => `missing MCP tool module: lib/mcp/${name}`),
...catalog.profileMissing.map((name) => `ChatGPT profile references missing MCP tool: ${name}`),
+ ...catalog.appOnlyMissing.map((name) => `ChatGPT app-only profile references missing MCP tool: ${name}`),
];
if (errors.length) { for (const error of errors) console.error(error); process.exit(1); }
const expected = renderMcpCatalogMarkdown(catalog);
diff --git a/scripts/lib/mcp-catalog.mjs b/scripts/lib/mcp-catalog.mjs
index c06a17ce..36abc091 100644
--- a/scripts/lib/mcp-catalog.mjs
+++ b/scripts/lib/mcp-catalog.mjs
@@ -1,7 +1,6 @@
import { readFileSync, readdirSync } from "node:fs";
import { join } from "node:path";
-export const APP_ONLY_MCP_TOOLS = new Set(["workflow_status", "render_mso_surface"]);
const TOOL_MODULE_NAME = /^tools-[A-Za-z0-9-]+\.ts$/;
const SCOPES = ["read", "write", "exec"];
@@ -43,16 +42,22 @@ export function collectMcpCatalog(root) {
for (const match of text.matchAll(/name:\s*"([^"]+)"[\s\S]*?scope:\s*"(read|write|exec)"/g)) all.set(match[1], match[2]);
}
- const appOnly = [...APP_ONLY_MCP_TOOLS].filter((name) => all.has(name)).sort();
- const model = [...all].filter(([name]) => !APP_ONLY_MCP_TOOLS.has(name));
+ const toolContract = read("lib/mcp/tool-contract.ts");
+ const appOnlyBlock = /CHATGPT_APP_ONLY_TOOL_NAMES\s*=\s*new Set\(\[([\s\S]*?)\]\s*as const\)/.exec(toolContract)?.[1] ?? "";
+ const appOnlyNames = [...appOnlyBlock.matchAll(/"([^"]+)"/g)].map((match) => match[1]);
+ const appOnlySet = new Set(appOnlyNames);
+ const appOnlyMissing = appOnlyNames.filter((name) => !all.has(name));
+ const appOnly = appOnlyNames.filter((name) => all.has(name)).sort();
+ const model = [...all].filter(([name]) => !appOnlySet.has(name));
const byScope = splitByScope(model);
- const toolContract = read("lib/mcp/tool-contract.ts");
- const profileBlock = /CHATGPT_TOOL_NAMES\s*=\s*new Set\(\[([\s\S]*?)\]\s*as const\)/.exec(toolContract)?.[1] ?? "";
- const chatgptNames = [...profileBlock.matchAll(/"([^"]+)"/g)].map((match) => match[1]);
- const profileMissing = chatgptNames.filter((name) => !all.has(name));
- const chatgptAppOnly = chatgptNames.filter((name) => APP_ONLY_MCP_TOOLS.has(name) && all.has(name)).sort();
- const chatgptModel = chatgptNames.filter((name) => !APP_ONLY_MCP_TOOLS.has(name) && all.has(name)).map((name) => [name, all.get(name)]);
+ // ChatGPT receives the complete MSO-owned generic model/operator catalog.
+ // The two compatibility bridges remain app-only and project-owned MCP names
+ // remain dynamic behind project_mcp_tools/project_mcp_call.
+ const chatgptNames = [...all.keys()];
+ const profileMissing = [];
+ const chatgptAppOnly = [...appOnly];
+ const chatgptModel = [...model];
const chatgptByScope = splitByScope(chatgptModel);
return {
@@ -69,7 +74,7 @@ export function collectMcpCatalog(root) {
modelCount: chatgptModel.length,
counts: countsFor(chatgptByScope),
},
- orphanModules, missingModules, profileMissing,
+ orphanModules, missingModules, profileMissing, appOnlyMissing,
};
}
@@ -78,7 +83,7 @@ export function renderMcpCatalogMarkdown(catalog) {
const c = catalog.chatgpt;
return `# Generated MCP catalog
-> **Generated file — do not edit manually.** Source of truth: \`lib/mcp/tools.ts\`, its registered \`tools-*\` modules, \`lib/mcp/toolset.ts\`, and the compact ChatGPT name set in \`lib/mcp/tool-contract.ts\`. Regenerate with \`node scripts/gen-mcp-catalog.mjs\`. The live deployed authority remains \`GET /mcp\`.
+> **Generated file — do not edit manually.** Source of truth: \`lib/mcp/tools.ts\`, its registered \`tools-*\` modules, \`lib/mcp/toolset.ts\`, and the ChatGPT app-only exclusions in \`lib/mcp/tool-contract.ts\`. Regenerate with \`node scripts/gen-mcp-catalog.mjs\`. The live deployed authority remains \`GET /mcp\`.
## Full MSO catalog
@@ -112,11 +117,11 @@ ${list(catalog.byScope.exec)}
${list(catalog.appOnly)}
-## ChatGPT static profile
+## ChatGPT model profile
<!-- mcp-chatgpt-profile: server=${catalog.serverVersion} version=${catalog.version} tools=${c.modelCount} read=${c.counts.read} write=${c.counts.write} exec=${c.counts.exec} app-only=${c.appOnly.length} total=${c.transportCount} -->
-The ChatGPT profile is a fail-closed static projection defined by \`CHATGPT_TOOL_NAMES\`. OAuth scope is still enforced independently; project-owned MCP tool names remain dynamic data behind the generic project bridge.
+The ChatGPT profile automatically projects the complete MSO-owned generic model/operator catalog. OAuth scope is still enforced independently; app-only compatibility bridges stay app-only, and project-owned MCP tool names remain dynamic data behind the generic project bridge.
| Fact | Current source value |
|---|---:|