Skip to main content

Remote MCP

Chronote exposes a remote MCP (Model Context Protocol) endpoint for AI assistants and agents that need meeting context or want to control live recordings.

The endpoint uses OAuth with Discord sign-in. Chronote checks the same meeting access rules used by the web portal before returning data, and write tools require separate consent scopes.

For an unattended agent that cannot complete a browser sign-in, see Service Accounts.

Endpoint​

Use the API MCP endpoint:

https://api.chronote.gg/mcp

Self-hosted deployments should use their own API base URL plus /mcp.

Remote MCP requires Discord OAuth and OAUTH_SECRET to be configured. Set MCP_PUBLIC_BASE_URL to the externally reachable API origin if it differs from the server's internal URL.

Available Tools​

  • list_servers: Lists Discord servers where the authenticated user can access Chronote data.
  • list_meetings: Lists recent meetings in a server, with optional channel, date, tag, and archived filters.
  • list_my_meetings: Lists meetings across servers for the authenticated user. It defaults to meetings the user attended in the past 7 days and can also list meetings the user can access. Use all, today, or past_7_days without startDate or endDate. Use range: "custom" when you need explicit date bounds. If the response includes nextCursor, pass it as cursor to fetch the next page.
  • get_meeting_summary: Returns notes and metadata for one accessible meeting.
  • get_meeting_transcript: Returns transcript text for one accessible meeting when that server allows transcript access. Use offset and maxChars to page through long transcripts.
  • start_meeting: Starts a Chronote recording from the authenticated user's current Discord voice channel. Optionally pass serverId, voiceChannelId, textChannelId, context, and tags.
  • stop_meeting: Stops the active Chronote meeting, either by serverId or by inferring the server from the authenticated user's current voice channel. Pass meetingId to guard against stopping the wrong meeting.
  • get_live_meeting_status: Returns status for an active meeting.
  • get_live_meeting_transcript: Returns currently available live transcript events when that server allows transcript access. Pass serverId so the request reaches the bot runtime that owns the meeting, and use afterEventId to page from the last event you saw.
  • get_meeting_control_request: Checks a queued start, stop, or live meeting request by requestId.

For follow-up fetch tools, use the list item's id field. Do not pass the UUID-style meetingId field.

Example flow:

  1. Call list_meetings or list_my_meetings.
  2. Take the returned id value, for example <server-id>#2026-05-08T01:06:47.307Z.
  3. Pass that id into get_meeting_summary or get_meeting_transcript.

For long transcripts, request a window at a time:

  1. Call get_meeting_transcript with id, optionally adding maxChars.
  2. If the response includes truncated: true, call it again with offset set to nextOffset.

Live Control Tips​

  • Join the Discord voice channel before calling start_meeting without explicit IDs.
  • If you are connected in multiple servers, pass serverId.
  • If the server does not have a default notes channel, pass textChannelId.
  • Write and live tools may return requestStatus: "pending" with a requestId while a bot worker completes the command. Call get_meeting_control_request with that requestId to check the result.
  • stop_meeting can be used by the meeting creator or members with meeting-management permissions.

Date Range Tips​

  • For preset My Meetings ranges, send range: "all", range: "today", or range: "past_7_days" and omit startDate and endDate.
  • For explicit windows, send range: "custom" with startDate, and optionally endDate.
  • Chronote rejects mixed inputs such as range: "past_7_days" with a manual startDate.

OAuth Scopes​

  • meetings:read: Required for server lists, meeting lists, and meeting summaries.
  • transcripts:read: Required in addition to meetings:read for transcript text.
  • meetings:start: Required to start recordings through MCP.
  • meetings:stop: Required to stop recordings through MCP.
  • get_meeting_control_request requires a valid MCP token and only returns requests created by the authenticated user.

If a tool returns insufficient_scope, your MCP client should follow the WWW-Authenticate challenge, run OAuth step-up authorization for the listed scopes, and retry the request. During step-up, Chronote may require a fresh browser authorization instead of refreshing the existing token so the new scopes are explicitly approved. If your client does not support step-up authorization, reconnect Chronote MCP and approve the read, transcript, start, and stop scopes you need.

OpenCode Example​

{
  "mcp": {
    "chronote": {
      "type": "remote",
      "url": "https://api.chronote.gg/mcp",
      "oauth": {}
    }
  }
}

When your MCP client connects, it opens a Chronote authorization flow in the browser. Sign in with Discord, review the requested scopes, and approve the client.

Access Rules​

Chronote only returns meeting data if the authenticated Discord user can access the meeting. Access follows the same model as My Meetings and Server Library:

  • Users who participated can access their indexed meetings when attendee access is enabled.
  • Other users need access to the voice channel and notes channel history.
  • Transcript access requires explicit transcript scope consent.
  • A server's Transcript access setting can still make completed and live transcripts unavailable, even when the token has transcripts:read.
  • Starting a live meeting requires the authenticated Discord user to already be in the target voice channel.
  • Stopping a meeting requires creator or meeting-management permissions.

Service account tokens run through these same rules, resolved against the bot they act as, plus the server and channel bounds recorded on the token.