Where do my transcripts go?

The server runs on your own computer and talks to your Wazomind account directly. Whatever you then ask your assistant about is handled by that assistant under its own privacy terms, so only connect it to one you already trust with your work.

MCP server

Ask Claude about your own calls

Q: Does it help during a live call?

No. MCP works by request and reply: the assistant asks, the server answers. Live coaching during a call needs your microphone and an on-screen overlay, which is what the Wazomind Chrome extension and Windows app do. The MCP server covers before and after the call.

Q: Can it change or delete my meetings?

No. Every tool is read-only. Nothing in the MCP server can edit a meeting, delete one, create a public share link, or spend your credits. That is enforced by a test, not just by intention.

Q: Does it cost anything?

No. It reads the library you already have and works on the free plan. Asking questions is rate-limited per hour, and higher on Pro.

WazomindIntegration

Your meeting history, available to your AI assistant. Connect Wazomind to Claude, Claude Code or Cursor, and ask what a client said three weeks ago, what you promised and never sent, or what to raise on the call you are about to join.

You:    What did Acme push back on last time, and what did I promise them?

Claude: Acme raised the annual price twice [Acme — renewal call, 2026-10-14].
        You owe them the annual pricing sheet, due Friday, and a seat count
        confirmation. Neither is marked done.

What it is good for

Before a call

A brief built from your own past meetings with the same people: where things stand, who owes what, which objections keep coming up.

After a call

The notes, the action items and the follow-up draft, in the same window where you are already writing the email.

Across months

Ask a question across every call you have recorded and get an answer that cites the meeting and the date it came from.

It does not help during the call. MCP works by request and reply — your assistant asks, the server answers. Live coaching needs your microphone and an overlay on screen, which is the job of the Chrome extension and the Windows app. This covers before and after.

Set it up

You need Node 22.6 or newer, and a Wazomind account. Open your meeting library, go to Settings, and copy your connection token from the Claude & other AI assistants card.

Claude Code

claude mcp add wazomind \
  --env WAZOMIND_TOKEN=your-token-here \
  -- npx -y wazomind-mcp

Claude Desktop or Cursor

Add this to claude_desktop_config.json (or Cursor's mcp.json), then restart:

{
  "mcpServers": {
    "wazomind": {
      "command": "npx",
      "args": ["-y", "wazomind-mcp"],
      "env": { "WAZOMIND_TOKEN": "your-token-here" }
    }
  }
}

Then ask your assistant something about a past meeting. It will show you the Wazomind tool it is about to use and wait for your approval the first time.

What your assistant can do

Tool	What it does
`list_meetings`	Recent meetings with date, length and how many action items are still open
`search_meetings`	Keyword search across every transcript, returning the lines that match
`get_meeting`	One meeting's notes, action items and follow-up draft
`ask_meetings`	A question answered from your transcripts, citing the meetings it used
`prep_call_brief`	A pre-call brief from your history with the people you are about to meet
`list_open_actions`	Every commitment still open across your recent meetings
`get_followup_draft`	The follow-up email Wazomind already wrote for a meeting

It can only read

Handing an AI assistant access to your call history is a real decision, so the server is built to be dull on purpose:

Nothing can be changed or deleted

Every tool is read-only. No tool edits a meeting, deletes one, or creates a public share link. A test fails the build if a write ever appears.

Nothing can be spent

Generating notes can cost a credit, so that is not exposed at all. Your assistant cannot spend your balance.

One account, your token

The server holds no credentials of its own. It reaches exactly the one account your token belongs to, and you can stop it by removing the token.

Whatever you then ask your assistant is handled by that assistant under its own privacy terms — so connect it to one you already trust with your work. The code is small and readable if you would rather check it yourself.

FAQ

What is an MCP server?

MCP, the Model Context Protocol, is an open standard for giving an AI assistant access to a tool or a source of data. An MCP server is the small program that exposes it. Install this one and Claude can read your Wazomind meeting library when you ask it to.

Does it help during a live call?

No. MCP is request and reply, so it cannot hear anything. Live answers during a call come from the Chrome extension or the Windows app, which run on your side of the call with no bot joining. The MCP server is for before and after.

Can it change or delete my meetings?

No. Every tool is read-only, and that is enforced by a test rather than left to good intentions. It cannot edit, delete, share or spend.

Does it cost anything?

No. It reads the library you already have, and it works on the free plan. Asking questions is rate-limited per hour, with a higher limit on Pro.

Which assistants work with it?

Anything that speaks MCP over stdio: Claude Desktop, Claude Code, Cursor, and a growing number of others. You need Node 22.6 or newer installed.

Why do I have to paste a token?

Because this is an early version. A proper sign-in button is the next thing to build here. Until then the token is how your assistant proves it is you, and you can revoke it by signing out.

What about ChatGPT?

Not yet. ChatGPT reads MCP servers over a different transport than the one this ships with, so it needs more work. The groundwork is in place.

Notetakers give you a transcript to go and read. This puts the answer in the window you are already working in.

Record a call first

The MCP server reads the meetings you have already recorded, so start there. Free to begin, no card.

Get Wazomind free