---
title: "Give your agent a terminal over MCP"
url: "https://docs.phux.sh/terminal-mcp"
source: "https://github.com/no-phux/website/blob/bffc0f5a462084993a7cfe1367f97d7ac533cf9f/src/site-docs/terminal-mcp.md"
projectVersion: "0.113.1"
protocolVersion: "1.0.0"
stability: "stable"
---

# Give your agent a terminal over MCP

phux mcp serves your local phux server as MCP tools on stdin and stdout. Register it with the agent the same way you register any other stdio server.

```sh
claude mcp add phux -- phux mcp
phux mcp --schema
```

With no arguments, `phux mcp` serves MCP on stdin and stdout for the phux server on this machine. It forwards every argument to the MCP companion. `--schema` prints each tool and its input schema. `--skill` prints the guide for those tools. `--help` and `--version` are the other discovery modes.

Each tool mirrors one `phux` command and is marked read-only or destructive the same way the server treats it. Tools that kill, detach, send a destructive signal, or approve something require `confirm: true`.

The line above is how Claude Code registers a stdio server. Plugins for Claude Code, pi, Oh My Pi, OpenCode, and Grok register it for you. The install commands are on [Coding agents](/agents#mcp).

This server can type into your panes. The hosted server at [phux.sh/mcp](https://phux.sh/mcp) only searches documentation. It cannot see or control a local terminal. Connect that one with `claude mcp add --transport http phux-docs https://phux.sh/mcp`.

## Next

- [Coding agents](/agents#mcp)
- [Run a command in a pane](/pane-command)