munim-computer-use

Computer Use MCP server that lets any coding agent use your computer, on macOS, Windows, and Linux.

macOSWindowsLinux

Overview

munim-computer-use is an open-source MCP server that lets any coding agent use your computer the way a person does. It reads the screen through accessibility trees, clicks and types in the background so your mouse stays yours, shows an agent pointer where it is working, zooms in on small text, and drives tabs in your signed-in Chrome. It is the Computer Use engine of MT Code, published on its own, and works with Claude Code, Codex, Cursor, MT Code, or any other MCP client with any model.

Installation

Terminal

claude mcp add munim-computer-use -- npx -y munim-computer-use

Usage

# Claude Code - the npm launcher fetches the signed binary on first run
claude mcp add munim-computer-use -- npx -y munim-computer-use

# Codex - ~/.codex/config.toml
[mcp_servers.munim-computer-use]
command = "npx"
args = ["-y", "munim-computer-use"]

# Cursor - .cursor/mcp.json
{ "mcpServers": { "munim-computer-use": { "command": "npx", "args": ["-y", "munim-computer-use"] } } }

# macOS only, once
munim-computer-use request-permissions

Requirements

An MCP client such as Claude Code, Codex, Cursor, or MT Code. On macOS, run munim-computer-use request-permissions once to grant Accessibility and Screen Recording. Prebuilt binaries cover macOS universal (Developer ID signed), Windows x64, and Linux x64 and arm64 (glibc 2.39+, X11); the npm launcher fetches the signed binary on first run. The Chrome extension is optional and only needed for browser tools.

Platform notes

macOS

Uses the accessibility API for element ids and SkyLight for background input. Requires Accessibility and Screen Recording permissions, granted once via request-permissions. Prebuilt universal binary is signed with a Developer ID.

Windows

Uses UI Automation for element ids and posted window messages for background input. Prebuilt x64 binary is available on Releases.

Linux

Uses AT-SPI and X11 with XTEST for background input. Native Wayland apps get element actions but not coordinate clicks.

Features

•Accessibility first - get_app_state returns the accessibility tree with stable element ids, so the agent presses the button instead of guessing at a coordinate
•Background control - Events are addressed to the target window, so there is no focus stealing and no hijacked mouse
•Agent pointer overlay - A soft lavender pointer shows where the agent is acting while your own cursor stays untouched
•Coordinates that land - Every screenshot and zoom carries its screen origin and pixels-per-point
•Your browser, your logins - The Chrome extension gives the agent its own labelled tab group in your signed-in Chrome
•Tab adoption - browser_use_tab takes over a tab you already have open in place, without moving, activating, or reloading it
•Model agnostic - No vision model is required for interaction, and local models work too
•32 tools - Byte-identical schemas across every supported platform
•Any MCP client - Works with Claude Code, Codex, Cursor, MT Code, and anything else that speaks MCP
•Apache-2.0 licensed with prebuilt signed binaries and an npm launcher

For the latest API reference, usage examples, and troubleshooting, visit the GitHub repository and npm package page.