Build what your team agreed: connect Meeting.ai to Claude Code

Let Claude Code read the meeting itself, and you stop pasting a summary into the prompt and hoping you remembered it right. Meeting.ai is the recorder: it keeps the notes and the transcript of your meetings. Once it is connected, Claude Code reads them while it works in your repository and writes the ADR, the test or the code from what was said.

Setup is two commands and one browser sign-in, with no API key to store. It works on meetings Meeting.ai has recorded for you: a call its notetaker joined, or a recording you made or uploaded.

Three cases from one sprint.

The decision record, written while people still remember the reasons. On Tuesday the team spent forty minutes choosing a job queue. By next month nobody will know why Redis lost. You type:

From Tuesday's architecture meeting in Meeting.ai, write an ADR in docs/adr/ using the existing template: context, the options we discussed, the decision, and the consequences. Quote the transcript where the decision was made.

Claude Code finds the meeting, reads the notes, searches the transcript and writes the file. The output blocks in this post are illustrations from invented meetings:

Wrote docs/adr/0014-job-queue-on-postgres.md

Decision: background jobs run on Postgres with SKIP LOCKED. No Redis.
Why: about 40 jobs a minute, and one less service to keep running.
Rejected: Redis Streams (a second datastore), SQS (local dev needs AWS).
Quoted: "We don't have the volume to justify Redis."

The customer's bug, as a failing test. On Thursday's call a customer explained how their invoice export shows the wrong day. Nobody wrote the steps down.

Find Thursday's call with Northwind in Meeting.ai. Extract the exact steps they followed in the invoice export and what they expected to see. Write a failing test that reproduces it. Do not fix it yet.

Claude Code pulls the steps from the transcript and adds test_export_uses_account_timezone. It fails: the customer in Auckland expected 1 October, and the export says 30 September. The steps in the test are the customer's own.

The call that is still running. The rate-limit review is about to start. You paste its invite link:

Send the Meeting Bot to this call: [invite link]

Claude Code asks before it uses the tool, unless your permission settings already allow it. The Meeting Bot, Meeting.ai's notetaker, joins the Zoom, Google Meet or Microsoft Teams call as a participant everyone sees. Twenty-five minutes in, before you open a branch, you ask:

What has this call decided so far about the rate limiter, and what is still open? Do not write code yet.

Decided so far:
- Limit per API key, not per IP address.
- Token bucket: 100 requests a minute, burst of 20.

Still open:
- Whether a 429 response carries a Retry-After header.

You start on the two settled points and leave the header alone.

How do you connect Meeting.ai to Claude Code?

Install the Meeting.ai plugin from inside a Claude Code session. If Meeting.ai is already a connector in your claude.ai account, skip the install: sign in to Claude Code with the same account, type /mcp and look for the server marked as coming from claude.ai.

/plugin marketplace add meeting-ai/memo-claude-plugin
/plugin install meeting-ai@meeting-ai

1. Review the plugin. The second command opens the plugin's details. Review it and choose a scope.
2. Sign in. Type /mcp, select the Meeting.ai server, press Enter and choose Authenticate. Your browser opens the Meeting.ai sign-in page. If it does not, copy the URL shown in the terminal and open it yourself.
3. Allow access. Pick the workspace under Starts in workspace, read what the app is allowed to do, and click Allow access. Back in Claude Code, the server's status changes to connected.

Test it with something you can check:

List my five most recent meetings in Meeting.ai, with their dates.

If the list matches the Meeting.ai app, you are done. If you have no meetings yet, record or upload one first.

The same install works from your shell. The plugin then loads the next time you start Claude Code.

claude plugin marketplace add meeting-ai/memo-claude-plugin
claude plugin install meeting-ai@meeting-ai

Other routes: the MCP server alone, or the memo CLI

The plugin is the hosted Meeting.ai MCP server plus five skills: an overview, setup, recording a meeting, editing notes, and Visual Notes (a Visual Note is a one-page drawing of a meeting). Two other routes lead to the same Meeting.ai account. Add the MCP server alone if you want only the tools. Use the memo CLI if you want shell commands that print JSON.

Add the MCP server without the plugin

Use this route or the plugin, not both: they reach the same hosted server with the same sign-in. Run this in your terminal, not inside a claude session:

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

1. Check the status. Run claude mcp list. The server shows ! Needs authentication, which is expected before you sign in.
2. Sign in. Start Claude Code and follow steps 2 and 3 of the plugin route. The server is listed as meeting-ai.

You can also sign in from the shell with claude mcp login meeting-ai.

Choose the scope

The default is local scope: private to you, current project only. Add --scope user for all your projects, or --scope project to share the server with your team. Claude Code then writes a .mcp.json file in the project root that you can commit:

{
  "mcpServers": {
    "meeting-ai": {
      "type": "http",
      "url": "https://mcp.meeting.ai/mcp"
    }
  }
}

The file holds only the URL. Claude Code asks each teammate to approve a project-scoped server before using it. Everyone signs in with their own Meeting.ai account, so each person sees only the meetings that account can access.

Use the memo CLI

memo is the Meeting.ai command-line tool. It gives Claude Code the same meetings as shell commands that print JSON:

curl -fsSL https://raw.githubusercontent.com/meeting-ai/memo-cli/main/install.sh | bash

Then open a new terminal, so your shell finds memo in ~/.local/bin, and run:

memo auth login
memo meetings search --keyword "webhook"
memo meetings get <id> --all

The installer supports macOS and Linux and verifies the release checksum. memo install-skill-md adds the skill file that teaches a coding agent the commands. See the Meeting.ai CLI page and the CLI guide.

Good to know

• Read the plugin before you install it. As of 14 October 2026 it is not in Anthropic's plugin directory, and Anthropic does not review third-party marketplaces. It installs from our public repository, which holds two JSON manifests and five Markdown skills, and no server code.
• What we ran ourselves. We ran the claude mcp add command ourselves, on Claude Code 2.1.292. We wrote the plugin install and the browser sign-in from Anthropic's documentation and our plugin's README and have not run them end to end, so tell us if a step looks different on your machine.
• Cost. Connecting, searching and reading need no extra purchase while your plan gives you access to the meeting. Sending the Meeting Bot to a call, Visual Notes and media generation use your Meeting.ai subscription or Coins, the credits in your Meeting.ai account.
• Read before you approve a write. Seven of the 20 tools only read. The rest edit notes, share or export meetings, send the Meeting Bot or generate media, and some spend Coins.
• Notes first, transcript second. The notes hold the decisions, and usually the action items, in a few sections. Transcripts are long and come back in pages, and Claude Code warns when a tool returns more than 10,000 tokens and limits output to 25,000 by default.
• Check what the meeting said. Ask for quotes and for a list of what was left open, because a decision said out loud is sometimes reversed later in the same call. Live notes are provisional and keep changing until the meeting ends.
• Keep transcripts out of the repository. Commit the decision, in an ADR or a PR description. Leave your colleagues' words out of the commit.

What to ask Claude Code

Ask for the artifact: the ADR, the test, the PR description. If Claude Code searches the repository when you meant a meeting, add "in Meeting.ai" to the prompt.

Build the feature from the planning meeting

Yesterday's planning call settled how account deletion works. You were there, and you still do not trust your memory of the third rule.

Find yesterday's planning meeting about account deletion in Meeting.ai. Read the notes, then search the transcript for the rules we agreed. Implement them in this repo with a test for each rule. Before you write code, show me the rules you found and anything the meeting left undecided.

Claude Code reports before it edits:

From "Account deletion planning" (13 October):
1. Deleting an account marks it deleted and signs out every session.
2. Data is purged after 30 days. Logging in before then restores it.
3. Invoices stay, detached from the user record.

Undecided in the meeting:
- Whether a team owner can delete a member's account.
  Two options were raised, no decision recorded.

Plan: app/accounts/deletion.py (new), app/jobs/purge.py (new),
tests in tests/accounts/test_deletion.py. Proceed?

Answer the open question, or take it back to the team, before you say yes.

A PR description that explains why

The reviewer was not in the call, and the diff does not say why the purge job waits 30 days.

Write the PR description for this branch. Explain why each change was made using the decisions from the planning meeting in Meeting.ai, and list what was decided in the meeting that this PR does not cover yet.

The list at the end is your next task.

One convention out of three meetings

Error handling came up in three architecture meetings, and the code base now handles errors in three ways.

What did we decide about error handling across the last three architecture meetings in Meeting.ai? Write it into CONTRIBUTING.md and name the meetings it came from.

Before it writes, ask it to show where the three meetings disagreed.

More things to ask

• A branch against the meeting. "Compare the diff on this branch with what we agreed in the API review meeting. List every point where the code differs and quote the transcript line for each."
• A plan file. "Turn Monday's planning meeting into a task list in docs/plan.md, ordered by dependency, with the owner named in the meeting."
• A review against your own rules. "Review this branch against the coding rules we agreed in last month's engineering meeting."
• The retro, checked against the code. "Which action items from the last retro touch this repository, and which are still not done?"
• Your standup update. "Write my standup update in three lines (done, next, blocked) from yesterday's commits and the decisions in yesterday's meetings."
• An onboarding guide. "Write an onboarding guide for a new engineer from the last five architecture meetings."
• One line from the transcript. "Search the transcript of Tuesday's design review for 'idempotency key' and quote what was said about it."

Outside the repository, the same connection transcribes a recording from a Google Drive link that anyone with the link can download (or one you upload in Meeting.ai), turns a meeting into a Visual Note, and stops the recording of a meeting you own.

Why can Claude Code not find my meetings?

Usually the sign-in was never finished, or Claude Code is looking in the wrong workspace.

How do you disconnect Meeting.ai from Claude Code?

Run claude mcp remove meeting-ai. Claude Code also deletes the sign-in it stored for that server. To sign out and keep the entry, run claude mcp logout meeting-ai or use Clear authentication in the /mcp menu. If you installed the plugin, uninstall the plugin instead.

To cut access from the Meeting.ai side, open Settings › MCP & CLI, find Claude Code under Connected MCP Client, and click Revoke. Access stops at once.

Start with the last decision your team made out loud

Install the Meeting.ai plugin, sign in, and ask Claude Code for the ADR. The Meeting.ai MCP page lists every tool.

Guides for other tools:

• How to connect Meeting.ai to Codex
• How to connect Meeting.ai to Claude, for the chat app on web, desktop and mobile
• How to connect Meeting.ai to ChatGPT
• Through the memo CLI: Cursor, OpenClaw and Hermes Agent