- Home
- AI Computer Use Agents
- open-computer-use
open-computer-use
What it is
open-computer-use is a free, MIT-licensed local MCP server for AI agents that inspect and operate desktop applications through accessibility interfaces. Its nine core tools handle app discovery, accessibility-tree and screenshot capture, clicking, scrolling, dragging, typing, key presses, secondary actions, and control values. The project runs entirely on the user’s machine and says core app actions use the accessibility layer without taking over the physical mouse and keyboard. Setup documentation covers Codex CLI and App, Claude Code, Gemini CLI, and opencode, with manual configuration for other MCP clients; a ZCode plugin installs the skill and connects an MCP server. Password managers are refused, while app launching, focus changes, and global input injection require explicit environment-variable gates. macOS 14 or later requires Accessibility and Screen Recording permissions. Windows and Linux use a signed-in desktop session, with Linux requiring AT-SPI2. Additional window-level tools are currently available on Windows, while macOS and Linux support is in progress. The project should be used only on authorized machines.
For and against
For
- Runs locally on the user’s machine
- Provides nine desktop interaction tools
- Supports several named MCP clients
- Includes explicit safety gates
- MIT-licensed and free
Against
- macOS requires Accessibility and Screen Recording permissions
- Linux requires AT-SPI2
- Additional window tools are currently Windows-only
- Requires a signed-in desktop session
The verdict
open-computer-use provides local accessibility-based desktop control for AI agents across three operating systems. Permission requirements, platform differences, and its high-privilege scope matter when planning deployment.
Who it suits
It suits developers and AI-agent users who need local desktop-app control through accessibility interfaces on macOS, Windows, or Linux.
Plans and prices
Open source
MIT-licensed · runs locally · macOS, Windows, Linux
Read 9 Oct 2026 · github.com
On the record
- Desktop actions
- Its nine core tools support app discovery, accessibility-tree and screenshot capture, clicking, scrolling, dragging, typing, key presses, secondary actions, and setting control values.github.com · read 9 Oct 2026
- Integrations
- The project documents setup commands for Codex CLI and App, Claude Code, Gemini CLI, and opencode, and supports manual configuration for other MCP clients.github.com · read 9 Oct 2026
- License
- The repository is distributed under the MIT license.github.com · read 9 Oct 2026
- Local operation
- The project says it runs entirely on the user's machine and uses the accessibility layer without taking over the real mouse and keyboard for its core app actions.github.com · read 9 Oct 2026
- Permissions
- On macOS 14 or later, the project requires Accessibility and Screen Recording permissions; Windows and Linux work in a signed-in desktop session, with Linux requiring AT-SPI2.github.com · read 9 Oct 2026
- Platform limitation
- The additional window-level tools are currently available on Windows, with macOS and Linux support in progress.github.com · read 9 Oct 2026
- Purpose
- Open Computer Use is a local MCP server that lets AI agents inspect and operate desktop apps through accessibility interfaces.github.com · read 9 Oct 2026
- Safety controls
- Password managers are refused, while app launching, focus changes, and global input injection require explicit environment-variable gates.github.com · read 9 Oct 2026
- Security policy
- The security policy says fixes are provided for the latest npm release and asks users to report suspected vulnerabilities through GitHub's private vulnerability reporting channel.github.com · read 9 Oct 2026
- Security scope
- The policy notes that the tool performs high-privilege accessibility and input automation in the local signed-in desktop session and should only be used on authorized machines.github.com · read 9 Oct 2026
- Support and audience
- The project provides installation and troubleshooting references and invites contributions; its security policy says fixes are delivered through the regular release process.github.com · read 9 Oct 2026
- ZCode plugin
- The repository includes a ZCode plugin that installs the skill and connects an MCP server.github.com · read 9 Oct 2026