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

# MCP quick start

> Connect an AI agent to Open Headers over the Model Context Protocol — rules, requests, and workflows as tools.

By the end of this page an MCP-capable agent is connected to your Open
Headers instance and can work with your rules, requests, and workflows
as tools. The MCP surface is off by default and permission-tiered — you
choose what agents may observe, write, or execute.

## Enable the MCP surface

In the desktop app or extension, turn it on under **Settings → MCP**
and mint an access token there. On a self-hosted server (daemon
stopped):

```sh theme={null}
ohd config set mcp.enabled true
```

The observe / write / execute / secrets tiers are separate switches,
all off by default — enable only what your agent needs.

## Connect an agent

The MCP endpoint is streamable HTTP at `/mcp` on the same bind as
everything else, authenticated with your token as a bearer:

```
http://127.0.0.1:8137/mcp
Authorization: Bearer <token>
```

For Claude Code:

```sh theme={null}
claude mcp add --transport http open-headers http://127.0.0.1:8137/mcp \
  --header "Authorization: Bearer <token>"
```

Any MCP client that speaks streamable HTTP connects the same way.
Browser-originated requests to `/mcp` are refused outright, and failed
token attempts feed the per-peer rate limit.

## What the agent can do

The tool set mirrors the CLI: list and toggle rules, switch
environments and workspaces, set variables, send saved requests, run
workflows — always in sync with what the UI shows. The full tool
reference is coming to the [API & MCP section](/api-mcp).