generated: '2026-08-05'
method: searched
status: published
source: https://support.read.ai/hc/en-us/articles/49381158409491-MCP-Server
description: >-
  Read AI operates a first-party remote MCP server that exposes a user's meeting
  reports — metadata, participants, summaries, chapter summaries, action items, key
  questions, topics, full transcripts, engagement metrics, and a recording download
  URL — to any MCP client. It is listed in the Anthropic connector directory and in
  the ChatGPT app directory. Open beta as of 2026-08.
server:
  name: read-ai
  transport: streamable-http
  url: https://api.read.ai/mcp
  authentication:
    type: oauth2.1
    dynamic_client_registration: true
    registration_endpoint: https://api.read.ai/oauth/register
    authorization_server: https://authn.read.ai/
    protected_resource_metadata: https://api.read.ai/.well-known/oauth-protected-resource/mcp
    scopes: [openid, profile, email, offline_access, mcp:execute, meeting:read]
  status: open-beta
directories:
  - name: Anthropic connector directory
    url: https://claude.ai/directory/connectors/read-ai
  - name: ChatGPT apps directory
    url: https://chatgpt.com/
    note: Listed as "Read AI" under Apps -> All apps -> Browse Apps (per provider docs).
tools:
  - name: list_meetings
    description: >-
      Browse and filter meeting history with date-range filtering and cursor
      pagination. Returns the same detailed fields as get_meeting_by_id for each
      meeting in the page.
    parameters:
      - name: start_datetime_gte
        description: Lower bound on meeting start time.
      - name: start_datetime_lte
        description: Upper bound on meeting start time.
      - name: limit
        description: Page size. Capped at 10 by the server.
      - name: cursor
        description: ID of the last meeting from the previous page.
      - name: expand
        description: Array of expandable fields to include.
    source: >-
      Named verbatim in Read AI's own published Claude Skills
      (skills/read-ai-customer-feedback-digest.md, skills/read-ai-sales-call-review.md)
      and described in the MCP Server help article.
  - name: get_meeting_by_id
    description: >-
      Retrieve comprehensive information about a specific meeting by its ULID —
      basic metadata (title, start/end times, platform, report URL), participants
      with attendance status, summary, chapter summaries, action items, key
      questions, topics, full transcript with speaker identification, metrics, and a
      URL to download the meeting recording as MP4.
    parameters:
      - name: id
        description: Meeting ULID.
      - name: expand
        description: >-
          Array of: summary, chapter_summaries, action_items, key_questions, topics,
          transcript, metrics, recording_download.
    source: >-
      Named verbatim in skills/read-ai-sales-call-review.md; described in the MCP
      Server help article as "Get Meeting by ID".
compatible_clients: [ChatGPT, Claude Desktop, Claude Web, Claude Code]
x-evidence:
  fetched: '2026-08-05'
  probe: 'POST https://api.read.ai/mcp {"jsonrpc":"2.0","id":1,"method":"initialize"}'
  http_status: 401
  response: >-
    WWW-Authenticate: Bearer error="invalid_token" ... resource_metadata=
    "https://api.read.ai/.well-known/oauth-protected-resource/mcp" (server: uvicorn)
  note: >-
    Live tools/list is OAuth-gated, so no anonymous inputSchema introspection is
    possible. The two tool names and their parameter names above are taken from the
    provider's OWN published skill bundles, not guessed. Full JSON Schemas require an
    authenticated tools/list.
  gated: true