// docs · mcp

Meetings, in your assistant.

MCP (the Model Context Protocol) is a standard way to give an AI assistant access to a tool or data source. Point Claude Desktop or Cursor at the Notabium MCP server and your assistant can list, read, search, and ask questions of your meetings, with no glue code.

The server

Notabium runs a remote MCP server. There is nothing to install: you give your client one URL and one API key.

# remote MCP server URL
https://api.notabium.com/v1/mcp

Authentication is the same Notabium API key used by the REST API. Create one on the API keys page, and pass it as a bearer token in the Authorization header. The key is scoped to your meetings, and like all keys it is read and ask only.

The config block

MCP clients take a small JSON config that names each server. For Notabium it is one entry with the URL and the auth header:

{
  "mcpServers": {
    "notabium": {
      "url": "https://api.notabium.com/v1/mcp",
      "headers": {
        "Authorization": "Bearer ntb_live_..."
      }
    }
  }
}

Replace ntb_live_... with your real key. That is the whole configuration; the same block works across clients that support remote MCP servers.

Add it to Claude Desktop

  1. Create an API key on the API keys page and copy it.
  2. Open Claude Desktop, go to Settings, then Developer, and choose Edit Config. This opens claude_desktop_config.json.
  3. Paste the notabium entry from the config block above into mcpServers (merge it with anything already there), and put your real key in the Authorization header.
  4. Save the file and restart Claude Desktop.
  5. Notabium now appears in the tools menu. Try "summarize my last meeting."

Add it to Cursor

  1. Create an API key on the API keys page and copy it.
  2. In Cursor, open Settings, then MCP (or Tools and Integrations), and choose Add a new MCP server. This edits mcp.json.
  3. Add the notabium entry from the config block above, with your real key in the Authorization header.
  4. Save. Cursor connects to the server and lists its tools; enable them if prompted.
  5. In the chat, ask the agent something like "search my meetings for the pricing discussion."

Any other client that supports remote (HTTP) MCP servers uses the same URL, header, and config shape.

Available tools

Once connected, your assistant can call four tools. It picks the right one from your request, so you rarely name them yourself.

ToolWhat it does
list_meetingsList your most recent meetings (id, title, platform, date, status, duration).
get_meetingFetch one meeting by id: title, date, transcript text, and its summary, next steps, and email draft.
search_meetingsSearch your meetings by title or transcript text and get matching snippets.
ask_meetingAsk a question about one meeting, answered over its transcript.

Example prompts

Talk to your assistant in plain language. A few that work well:

Summarize my last meeting.
What did we decide about pricing across my calls this week?
Find the meeting where we talked about the launch date and pull the action items.
Draft a follow-up email from my call with Acme.

Behind the scenes the assistant calls list_meetings or search_meetings to find the right calls, then get_meeting or ask_meeting to read and answer. Everything stays scoped to your meetings.

Want the raw endpoints instead? They are all here.
REST API reference Tutorials