Reviewed by Jonathan West · Updated Oct 7, 2026

Granola MCP Setup Guide for Claude, ChatGPT, and Cursor

How to link Granola meeting notes to external AI assistants using the official Model Context Protocol server.

Reviewed by Jonathan West · Updated Oct 7, 2026

Granola provides an official Model Context Protocol (MCP) server that lets Claude, ChatGPT, and Cursor query meeting notes directly.

At Layer3Labs, we build workflow automations and CRM integrations for small and mid-sized teams, and every rollout we do starts by checking which data a connector can read.

Launched on February 4, 2026, the server runs on Streamable HTTP transport and uses OAuth authentication to grant assistants controlled access to meeting summaries and transcripts.


How to Connect the Granola MCP Server to Claude, ChatGPT and Cursor

Connecting Granola's Model Context Protocol (MCP) server to Claude, ChatGPT, or Cursor requires adding the connector in your client settings and completing a browser-based OAuth authorization.

Granola uses Dynamic Client Registration (DCR) for authentication, which eliminates the need to create a client ID or copy a client secret. Browser-based OAuth handles credentials automatically during setup.

If your company signs in through an identity provider, Granola also supports Enterprise-Managed Authorization (EMA). Granola lists Claude with Okta as a tested pair, using issuer https://mcp-auth.granola.ai and scope mcp.

  • Claude Desktop or Web: Open Claude, navigate to Settings, select Connectors, search for Granola, click Connect, and authenticate in your browser window.
  • ChatGPT: Open Settings, search for Granola, select Connect, and authenticate. You can then reference @Granola directly inside chat prompts.
  • Claude Code: In your terminal, run claude mcp add granola --transport http https://mcp.granola.ai/mcp, then type /mcp and choose Authenticate.
  • Cursor and custom MCP clients: Add a custom connection using Streamable HTTP transport pointing to the public endpoint https://mcp.granola.ai/mcp.
Granola MCP does not use static API keys or service accounts. Every client connection relies on OAuth authentication tied to an active user session.

Plan Scopes Across Basic, Business and Enterprise

The data accessible through Granola's MCP server depends on your subscription tier, which determines whether an assistant can read raw transcripts or search shared team folders.

On the free Basic plan ($0 per user per month), the MCP server can access only your personal notes from the last 30 days. Notes older than 30 days remain saved on Granola servers, but the MCP server cannot query them until you upgrade.

As of October 2026, the Business plan costs $14 per user per month on Granola's pricing page and adds unlimited meeting history and public workspace notes. Enterprise is $35 per user per month on the pricing page, and Granola's billing documentation words it as from $35. On Enterprise, an admin decides which scopes members can use.

Granola's original launch announcement on February 4, 2026 stated that users needed a paid subscription to use the Claude and ChatGPT connectors. Current Granola MCP documentation confirms that the Basic plan supports personal notes from the last 30 days.

  • Basic (Free): Personal notes from the last 30 days only. Folder inspection and raw transcript retrieval are unavailable.
  • Business ($14 per user per month): Personal notes, public workspace notes, unlimited meeting history, folder listings, and raw transcripts.
  • Enterprise ($35 per user per month): Administrator-configured scopes covering personal notes, public notes, or neither, plus an independent toggle for meeting transcripts.

The Six Tools in the Granola MCP Server

Granola exposes six distinct tools through its MCP server to let external AI assistants browse, retrieve, and chat with meeting records.

Four tools are available across all plans, while two tools require a paid subscription.

  • query_granola_meetings: Runs natural-language queries across meetings in your active workspace using Granola Chat capabilities.
  • list_meeting_folders: Lists your folders (paid plans only).
  • list_meetings: Lists your meetings so the assistant can pick which ones to open.
  • get_meetings: Fetches specific meeting details, including private notes and AI-generated summaries.
  • get_meeting_transcript: Retrieves the full, verbatim transcript of a selected meeting (paid plans only).
  • get_account_info: Returns details of the signed-in Granola account.
Summarized meeting notes give quick context, while raw transcripts consume more context window tokens. Use get_meetings for summaries and reserve get_meeting_transcript for exact quotes.

Prompt Workflows for Connected AI Assistants

Once Granola's MCP server is connected, an external assistant can search across conversations to summarize decisions or draft follow-up material.

The folder and transcript prompts below need a paid plan.

  • Weekly task audit: "Check my Granola meetings from last week and list all action items assigned to me, grouped by project."
  • Follow-up draft: "Retrieve the meeting transcript from yesterday's call with Acme Corp and draft a follow-up email confirming our deliverables."
  • Cross-meeting theme analysis: "Search my recent customer interviews for mentions of pricing objections and summarize the common themes."
  • Folder review: "List the meetings in my Product Feedback folder and extract the top three feature requests discussed this month."
  • Workspace check: "Which Granola workspace am I connected to?"

Rate Limits and Enterprise Administrative Controls

Granola's MCP rate limit currently averages about 100 requests per minute across all tools, and Granola says the limit may change.

The MCP server strictly follows the active workspace selected in your Granola desktop application. It does not aggregate notes across multiple workspaces, so you must switch workspaces in the desktop app to query different accounts.

According to Granola's MCP overview page, MCP access is turned off by default for Enterprise workspaces. Workspace administrators manage access under Settings > Workspace > General > Apps & connectors > MCP access.

Enterprise administrators can enable personal notes, public notes, both, or disable MCP access entirely. Admins can also toggle the separate Meeting transcripts switch to prevent raw conversation transcripts from leaving the workspace.


Architectural Differences Between MCP and the REST API

Choosing between Granola's MCP server and the Granola REST API depends on whether you need on-demand assistant retrieval or automated backend pipeline processing.

The MCP server connects conversational assistants like Claude and ChatGPT to your notes during interactive work sessions. It operates entirely on user-delegated OAuth permissions.

The Granola REST API (https://public-api.granola.ai/v1/notes) is designed for background jobs, CRM synchronizations, and server-side automation. API keys need a Business or Enterprise plan, as our Granola API guide explains. The API allows bursts of 25 requests per 5 seconds and a sustained 5 requests per second (300 per minute).

Granola MCP is not built for headless background synchronizations or automated cron jobs because it does not support static API keys or service accounts. Teams requiring automated data ingestion should use the REST API and webhooks instead.

If Granola introduces service account authentication or headless token provisioning for its MCP server, automated backend agents could use MCP without interactive browser OAuth logins.


Data Privacy Checks and Audit Logging

Connecting Granola's MCP server transmits meeting summaries and transcripts across external infrastructure operated by Anthropic, OpenAI, or Cursor.

Granola's security FAQ says notes are stored on Amazon Web Services (AWS) servers in the United States. Its security page says data is encrypted at rest and sent over TLS. Granola holds SOC 2 Type II certification.

Users on Basic and Business plans can opt out of model training in application preferences. Enterprise accounts have model training disabled by default across the entire workspace.

If your meetings include patient information, check the BAA scope first in our guide to Granola HIPAA compliance. Granola offers a Business Associate Agreement (BAA) only on the Enterprise plan. The BAA covers the desktop and laptop apps only, so a note an assistant reads through MCP falls outside it.

Enterprise administrators can monitor tool usage using the Granola Audit API. The Audit API tracks mcp.tool_used events alongside authentication and transcript view events.

Check which workspace is active in Granola, then add the Granola MCP connector in Claude and ask it to list your meetings from last week.

Frequently Asked Questions

  • Yes, Granola launched an official Model Context Protocol (MCP) server on February 4, 2026. It allows assistants like Claude, ChatGPT, and Cursor to query meeting notes using Streamable HTTP transport and OAuth authentication.
  • In Claude, open Settings, select Connectors, search for Granola, click Connect, and authenticate in your browser. In ChatGPT, open Settings, search for Granola, and connect. For Claude Code, run claude mcp add granola --transport http https://mcp.granola.ai/mcp in your terminal.
  • Yes, the MCP server works on Granola's free Basic plan. On the Basic plan, the server can only access personal notes created within the last 30 days, while Business and Enterprise plans provide unlimited history, folder access, and raw transcripts.
  • Yes, once you open ChatGPT Settings, search for Granola and connect it, you can query your meeting records by referencing @Granola in your chat prompts.
  • Yes, but only on paid Business and Enterprise plans. The get_meeting_transcript tool exposes full raw transcripts, while the free Basic plan restricts MCP access to meeting summaries and personal notes.

Audit Your AI Note-Taking and MCP Architecture

Layer3Labs helps organizations deploy AI assistants and MCP connectors without leaking confidential meeting records. We audit your workspace permissions, configure data retention policies, and establish secure AI workflows.

Book a Consultation