Sponsor

Open Computer Use

Open source computer use MCP server for macOS, Linux and Windows

Open Computer Use preview

About

Open Computer Use is an MIT licensed project by iFurySt (Leo) that wraps computer use as an MCP server. It reads app state through accessibility APIs instead of only screenshots, so agents can click, type and press keys in real apps without taking over your whole screen. It has passed 2,700 GitHub stars.

## What it does The server exposes tools such as `list_apps`, `get_app_state` and `press_key`, so an agent can inspect a running app, find an element by index and act on it. On macOS it also supports Locked Use: it can unlock a locked Mac temporarily so the agent keeps working, with a full screen shield in place. ## Install `npm i -g open-computer-use`, then run `ocu` once on macOS to grant Accessibility and Screen Recording. Wire it into your agent with `ocu install-claude-mcp`, `ocu install-codex-mcp`, `ocu install-gemini-mcp` or `ocu install-opencode-mcp`, or add `{"command": "open-computer-use", "args": ["mcp"]}` to any MCP config. ## Extras There is a companion skill (`npx skills add iFurySt/open-computer-use --skill open-computer-use`), a `doctor` command for permissions, and a `call` command to test single tools or sequences from the shell.

Features

  • Computer use over MCP on macOS, Linux and Windows
  • Accessibility based app state, not just screenshots
  • One command install for Claude Code, Codex, Gemini CLI and opencode
  • Locked Use on macOS with a protective screen shield
  • Companion agent skill
  • CLI to test tools and sequences

Install

Related skills

All skills
All articles

More MCP Servers

View all

Sourcery MCP

Let your coding agent request Sourcery reviews and fix the comments and security findings itself

anon.li

Let your agent manage private email aliases and encrypted file drops over OAuth MCP

ImageSVG

Turn images in ChatGPT into editable SVG through a remote MCP server