MCP reference

Mico's MCP server runs in Mico's cloud and reads the meetings you sync. It speaks MCP protocol 2025-03-26 over Streamable HTTP. Every tool is read-only and idempotent. It requires a plan that includes MCP (Flow), and the server checks the plan on every request. To connect a client, see MCP for agents.

Endpoint

https://<project>.supabase.co/functions/v1/mcp

The exact URL is shown in Settings → Developers → MCP.

• Every request is a POST with one JSON-RPC 2.0 message, and the response is JSON. Batches are not supported.
• The server is stateless. There is no session, and GET (the optional event stream) returns 405.
• Supported methods: initialize, ping, tools/list, and tools/call. Notifications are accepted with 202 and no body.
• The server declares only the tools capability. It has no resources or prompts.

Authentication

Send Authorization: Bearer <token> with one of these tokens:

For OAuth clients, the server publishes its metadata at /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server under the endpoint URL. It supports dynamic client registration (/register), the authorization code flow with PKCE S256 (/authorize and /token), public clients without a secret, and the mico:read scope. Refresh tokens last 30 days and rotate on every use.

On the authorization page you prove who you are with a personal access token. Revoking that token also revokes every OAuth connection it authorized.

A missing, unknown, expired, or revoked token returns 401 with a WWW-Authenticate header that points to the protected-resource metadata.

Tools

Out-of-range limit and offset values are clamped, not rejected.

list_meetings

Returns meetings, newest first, and pagination. pagination.total is null.

get_meeting

Accepts required string meeting_id. Returns the meeting metadata (id, title, kind, status, created_at, updated_at, started_at, ended_at, series_id, timezone, language), note, summaries, participants, and action_items. Transcript words are intentionally separate.

Each note or summary carries its text in markdown, as plain text with one line per paragraph or heading.

get_meeting_transcript

The result includes meeting_id, text, words, and pagination. pagination.total is the number of words in the transcript, and pagination.next_offset is null on the last page. If a meeting has transcript text but no word timings, the first page returns that text with an empty words list.

get_recurring_meeting_history

Accepts required string meeting_id plus optional integer limit and offset. The limit defaults to 20 and is clamped to 1..200; the offset defaults to 0. Returns meetings from the same recurring series, newest first, and pagination. A meeting outside a series returns an empty list.

search_meetings

Returns meetings, newest first, and pagination. pagination.total counts the matches.

list_pending_action_items

Accepts optional integer limit (default 20, clamped to 1..200) and offset (default 0). Returns action_items whose status is todo or in_progress, newest first, each with the meeting_id and meeting_title it came from, and pagination.

search_shared_notes

Searches notes other people share with you, including everything shared with your teams. Returns results, each with share_id, title, team, published_at, and a snippet. Notes and summaries only, never transcripts.

get_shared_note

Accepts required string share_id, as returned by search_shared_notes. Returns share_id, title, team, published_at, text (up to 12,000 characters), and truncated.

Results and errors

Each tool returns one text content item that contains JSON. Paginated results include a pagination object with offset, limit, returned, total, and next_offset.

A tool that cannot answer returns a normal result with isError: true and an error message in the JSON. That covers a missing required argument, a meeting that does not exist or is not yours (meeting 'ID' not found), and a failure while reading the data.

Protocol errors use JSON-RPC codes:

HTTP errors arrive instead of a JSON-RPC response: