AI assistants (MCP)

Connect with an API token

For Claude Code, Cursor and scripts — a token you create and paste once.

Make a token

  1. Profile → API tokens → New token.
  2. Name it after where it will live ("Claude Code on my laptop").
  3. Tick what it may do. Reading is ticked by default; writing is deliberate.
  4. Choose all the projects you can see, or only some.
  5. Choose how long it lasts: 30, 90, 180 or 365 days.

The token is shown once. It starts with cet_.

Claude Code

The dialog gives you the command with the token already in it:

claude mcp add --transport http cettara https://cettara.com/api/mcp --header "Authorization: Bearer cet_your_token"

Claude Desktop

Settings → Developer → Edit config:

{
  "mcpServers": {
    "cettara": {
      "type": "http",
      "url": "https://cettara.com/api/mcp",
      "headers": { "Authorization": "Bearer cet_your_token" }
    }
  }
}

Cursor

~/.cursor/mcp.json, or .cursor/mcp.json in the project:

{
  "mcpServers": {
    "cettara": {
      "url": "https://cettara.com/api/mcp",
      "headers": { "Authorization": "Bearer cet_your_token" }
    }
  }
}

VS Code

.vscode/mcp.json in the project — note it says servers, not mcpServers:

{
  "servers": {
    "cettara": {
      "type": "http",
      "url": "https://cettara.com/api/mcp",
      "headers": { "Authorization": "Bearer cet_your_token" }
    }
  }
}

Anything else

Every client speaks the same protocol to the same address:

  • Server URL: https://cettara.com/api/mcp
  • Transport: HTTP (streamable)
  • Header: Authorization: Bearer cet_your_token

The token dialog has a tab for each of these with your token already filled in.

Looking after it

  • A token acts as you. Anyone who has it can do what it allows.
  • It stops working when it expires, when you revoke it, when you leave the organization, or when the plan is no longer Business.
  • Last used is shown in the list, so an unused token is easy to spot and revoke.
  • Up to 20 active tokens each.

If a token leaks

Revoke it under Profile → API tokens. It stops at the next call — there is no delay. Then make a new one; nothing else needs changing.