Skip to content
Open
Show file tree
Hide file tree
Changes from 14 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 7 additions & 5 deletions .agents/skills/external-agents/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
5 changes: 5 additions & 0 deletions .changeset/mcp-tool-search-not-served.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
---
"@agent-native/core": patch
---

MCP `tool-search` now names the closest actions an app's own agent can run that the connection can't call. They come back under `notServed`, up to three with a short description each, with a message to hand the task to `ask_app`, so a capability kept off MCP no longer reads as one the app lacks. Only a signed-in person who may call `ask_app` gets them, and never for a menu search or a `readOnlyOnly` one. Denied, framework and turn-ending actions stay unnamed, as does anything the caller's OAuth scope or the app's discovery checks hide. When those checks fail or run past the search's deadline, the search comes back without `notServed`. `externalAgents.writes: "ask_app_only"`, set or implied by `authenticatedReads: "auto"`, is deprecated and warns once at startup: decide each write on its action instead, with `mcpTool: false` to keep it in the app.
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -250,11 +250,13 @@ outside the compact MCP Apps catalog. Apps can set
`externalAgents.authenticatedReads: "auto"` to automatically advertise every
action that is GET + `readOnly` + `publicAgent.requiresAuth`, instead of
maintaining a long connector allow-list. This policy never includes writes by
default; use `writes: "ask_app_only"` to keep mutations behind `ask_app`, and
use `denyActions` for defense-in-depth vetoes. `tool-search` remains available
for discovery, but a searched action still needs to be in the connector
catalog, included by the authenticated-read policy, or enabled by the explicit
full-catalog opt-in before `tools/call` can execute it. Do not rely on
default. Set `writes: "allowlisted"` to serve the writes your actions choose to
serve; `writes: "ask_app_only"`, the default with automatic reads, is
deprecated and warns at startup, so keep a write in the app with
`mcpTool: false` instead. Use `denyActions` for defense-in-depth vetoes.
`tool-search` finds only the tools the connection is served, listed or
`mcpTool: "search"` ones that `call-tool` runs. It names matching actions
outside them under `notServed`, for the caller to hand to `ask_app`. Do not rely on
action-specific `mcpApp` resources appearing in ChatGPT/Claude discovery by
default; use `open_app` for the
first-class app embed path. If a specific
Expand Down
2 changes: 1 addition & 1 deletion packages/core/docs/content/actions-access-control.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,7 @@ export default defineAction({
});
```

This is the same catalog as the plugin's `mcp: { connectorCatalog: [...] }` list, declared on the action instead of in a separate array that has to be kept in sync as actions are renamed. Both forms work, and an app can move over one action at a time. Everything left out stays discoverable through `tool-search`.
This is the same catalog as the plugin's `mcp: { connectorCatalog: [...] }` list, declared on the action instead of in a separate array that has to be kept in sync as actions are renamed. Both forms work, and an app can move over one action at a time. An action left out isn't on outside agents' default catalog. `mcpTool: "search"` serves one without listing it: `tool-search` finds it and `call-tool` runs it. For any other action your agent can call, `tool-search` names close matches as not served, so a signed-in outside agent can hand that task to `ask_app`.

Pairing the two flags the other way — `agentTool: false` with `mcpTool: true` — makes an action **MCP-only**: outside agents get it, your own app's agent never sees it. Use it for the export, sync, or handoff an outside tool needs and your own agent has no reason to call.

Expand Down
Loading
Loading