> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tai42.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Cursor

> Connect Cursor to a running server.

Connect Cursor to a running server with an MCP server config entry.

Cursor reads MCP servers from an `mcp.json` file and lists their tools in the editor. Add a `tai` entry — a launch command for a stdio server, or a URL for an HTTP one.

## Add a config entry

Put the entry in `~/.cursor/mcp.json` for all projects, or `.cursor/mcp.json` in a project for that project only. Add a `tai` server under `mcpServers`.

<CodeGroup>
  ```json stdio theme={null}
  {
    "mcpServers": {
      "tai": {
        "command": "tai",
        "args": ["serve", "--transport", "stdio", "--manifest-path", "/path/to/manifest.yml"]
      }
    }
  }
  ```

  ```json http theme={null}
  {
    "mcpServers": {
      "tai": {
        "url": "http://localhost:8000/mcp"
      }
    }
  }
  ```
</CodeGroup>

Use an absolute path to `tai` and to the manifest for the stdio entry. For the HTTP entry, run the server yourself (`tai serve --transport http --port 8000`) and point the URL at it.

<Note>
  A server behind [access control](/guides/access-control) requires an API key. For the stdio entry, set `TAI_API_KEY` in an `env` block on the server; for the HTTP entry, supply it in the client's request headers.
</Note>

## See also

* [MCP clients](/integrations/mcp-clients) — the transport options in depth.
* [Deploy](/guides/deploy) — running the server over HTTP.
* [Expose a private tool set](/guides/private-tool-set) — scope Cursor to a tool subset.