Granola MCP, how to put your meetings inside Claude Code, Cursor, and ChatGPT

One terminal command puts every meeting you've recorded in Granola inside Claude Code. The same server URL works in Cursor, and Claude.ai and ChatGPT pick it up from their connector menus. Granola shipped the MCP server on February 4, 2026, and most of what's written about it since is a directory listing with the URL pasted in. This is the working version. The exact setup for each app, what the free plan does and doesn't reach, and the prompts that make it worth the two minutes.

We run Claude Code every day and Granola on our calls, so the notes were already sitting there. Every setup step below was checked against Granola's MCP launch post and its MCP help doc on September 23, 2026. If Granola changes a menu name, their docs win.

What Granola MCP does

MCP (Model Context Protocol) is the open standard AI apps use to plug into outside data. Granola's own description is that it works "a bit like USB-C for AI apps." Granola runs an MCP server at a public URL. Once you connect an AI app to it and sign in, that app can search your meetings, read the notes, pull action items, and, on paid plans, read the raw transcript.

So your meeting history stops being a separate app you tab into. It becomes context the assistant can reach for on its own. Ask Claude Code to fix the bug the client described on Tuesday's call and it can go read Tuesday's call. That's the whole pitch, and it's a good one.

Which Granola plan you need

The Granola pricing page lists "MCP integration in all your apps" on the Business card, and its plan comparison table gives Basic MCP access to personal notes from the last 30 days. Granola's launch post and help doc fill in the rest. Basic connects with a narrower reach, and the paid plans open full history and transcripts.

Basic is a real free plan with unlimited notes and a 30-day history window, and it also includes AI chat across meetings and the option to opt out of model training at any time. For MCP, that means Basic is a fine test drive. It isn't where the value is. The prompts worth running mostly look back further than a month or need the transcript. Our Granola pricing breakdown has the full plan comparison.

Set up Granola MCP in Claude Code

Claude Code is the one client with a documented command in Granola's help center. Here it is, then the full sequence.

claude mcp add granola --transport http https://mcp.granola.ai/mcp

1. Run claude in your terminal once so Claude Code is installed and signed in.
2. Add the server with the command above.
3. Open a new terminal tab and run claude again.
4. Type /mcp, select granola, and choose Authenticate. A browser window opens for the Granola sign-in.
5. Ask Claude about your meetings.

That's the whole install. No config file to hand-edit, no API token to copy out of a settings page. If /mcp lists granola but Claude says it can't find meetings, the sign-in didn't finish. Run Authenticate again and watch for the browser tab.

Set up Granola MCP in Cursor

Granola's instructions for Cursor are one sentence. Add the server in Cursor's MCP configuration, then authenticate through the OAuth flow. Cursor's MCP documentation fills in the rest.

1. Open ~/.cursor/mcp.json for every project, or .cursor/mcp.json inside one repo if you only want it there.
2. Add the Granola server by URL.
3. Save, then find Granola in Cursor's MCP settings and enable it.
4. Complete the Granola sign-in in the browser window that opens.

{
  "mcpServers": {
    "granola": {
      "url": "https://mcp.granola.ai/mcp"
    }
  }
}

If you already have other servers in that file, add the granola block inside the existing mcpServers object instead of pasting a second one.

Connect Granola to Claude.ai and ChatGPT

The chat apps don't need a URL at all. Granola is listed in both connector directories.

Claude.ai and Claude desktop

1. Open Settings, then Connectors.
2. Search for Granola and connect it.
3. Sign in to Granola in the browser window.
4. Enable the Granola connector in the chat where you want it.

ChatGPT

1. Open Settings, then Apps.
2. Search for Granola and connect it.
3. Sign in to Granola in the browser window.
4. Mention @Granola in a chat, or just ask about a meeting by name.

Granola's client guide also lists Microsoft Copilot, Grok, Windsurf, and VS Code. Anything that supports MCP over Streamable HTTP with OAuth can connect using the same URL. That covers Codex too, which Granola files under other AI tools rather than giving it its own steps.

What the server can actually do

Granola's help doc lists six tools. Your AI app picks which one to call, so you never type these names, but knowing them tells you what's possible.

Notice what's missing. There's no tool that writes back to Granola. The server reads your meetings and hands them to the assistant. Anything that gets created, a ticket, an email draft, a code change, happens on the assistant's side.

8 prompts worth running

These map to the use cases Granola names in its own posts (turning discussed bugs into tickets, drafting call summaries, updating a board from standup notes, building proposals from discovery calls) plus a few that only make sense inside a code editor. Swap in your own client and project names.

1. "What did we agree on in the Acme kickoff, and what did I say I'd send them?" The basic memory check. Works on Basic if the kickoff was in the last 30 days.
2. "Pull every action item assigned to me from this week's meetings and group them by client." Faster than opening five notes one at a time.
3. "Draft a follow-up email to Dana from yesterday's call. Confirm the timeline and the two open questions." The assistant reads the note, you edit the draft.
4. "Read Tuesday's call with the client about the checkout bug, then find the code that handles that flow and propose a fix." This is the Claude Code and Cursor use case. The bug report is the meeting.
5. "Turn the bugs we discussed in today's standup into GitHub issues with repro steps." Pair it with a GitHub or Linear connection and the notes become tickets.
6. "Find every time a customer mentioned pricing in the last quarter and quote the exact words." Needs Business, since it reaches past 30 days and leans on transcripts.
7. "Before my 2pm, summarize every past meeting with this person and what's still open." Call prep in one message.
8. "Write a project brief for the new landing page from the three discovery calls in my Website folder." Folders are a paid-plan tool, and this is where they pay off.

Limits and gotchas

• Basic stops at 30 days and has no transcripts. The connection works, but the prompts that need history or exact quotes come back thin. Business at $14 per user per month is the fix.
• Sign-in is per app. Connecting Claude Code doesn't connect Cursor. Each client runs its own OAuth sign-in with Granola.
• You only see notes you own. Notes in a team space folder aren't reachable through MCP unless you own them individually.
• One workspace at a time. The server only returns notes from the active workspace and doesn't combine several.
• Enterprise starts switched off. If you're on a company plan and can't find Granola in your connectors, an admin has to enable MCP first.

On privacy, two layers matter. Inside Granola, Basic lets you opt out of model training any time, Enterprise sets that opt-out for the whole team, and Granola says the third-party AI providers it uses are contractually barred from training on your data. Outside Granola, remember what MCP does. The notes your assistant pulls get sent to that assistant's model. If a client call shouldn't touch ChatGPT, don't ask ChatGPT about it.

Who should set this up

• Developers who take client or product calls. Claude Code or Cursor reading the call where the requirement was described is the best use of this feature by a distance.
• Consultants and founders who already live in Claude or ChatGPT. Follow-ups, call prep, and weekly roll-ups stop requiring a copy-paste.
• Anyone already paying for Granola Business. It's included and it takes two minutes, so turn it on.

If you take three calls a week and read each note once, skip it. MCP pays off when there's a pile of meetings to search, not a handful. For the case for Granola itself, read our Granola AI review. If you're stacking this into a bigger Claude Code setup, our Claude Code Skills post covers how Skills and MCP servers fit together.

The verdict