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

# MCP Server

> Connect a coding agent to Monte

The Monte MCP server wraps the [CLI](https://docs.trymonte.ai/cli.md) for a coding agent. The agent runs the same commands, with the same arguments, rules, and refusals.

## Set Up the Server

`monte mcp` prints the server URL. Add the server to the agent with that URL, then restart the agent.

```bash theme={null}
monte mcp
```

> [tab] Claude Code
>
> ```bash theme={null}
> claude mcp add --transport http monte <url>
> ```

> [tab] Cursor
>
> Add the server to `~/.cursor/mcp.json`, or to `.cursor/mcp.json` in a project.
>
> ```json theme={null}
> {
>   "mcpServers": {
>     "monte": {
>       "url": "<url>"
>     }
>   }
> }
> ```

> [tab] Codex
>
> Add the server to `~/.codex/config.toml`.
>
> ```toml theme={null}
> [mcp_servers.monte]
> url = "<url>"
> ```

## Authenticate

On its first call, the agent opens a browser to sign in with the same account as `monte login`. The agent sees only the records of the teams that the account belongs to.

## Control What the Agent Changes

A command that changes state starts a job that uses GPU time, writes a record, or changes a live deployment.

> [!WARNING]
>
> Configure the agent to ask before it runs a command that changes state. In Claude Code, leave the `monte` server out of the allow list in `.claude/settings.json`. In Cursor and Codex, keep approval on for the `monte` server.

The CLI rules apply. A training run refuses to start until every eval has a baseline, and a held-out eval runs only when the call names it.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.