Open Computer Use MCP
open-computer-use
An open MCP server for desktop control: the agent reads app state and clicks, types and scrolls on macOS, Linux and Windows
- 2,940 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Drives the real desktop session: clicks, types and can change data in apps
- Requires Accessibility and Screen Recording permissions
Install
- Works in
Manual install
npm i -g open-computer-useInstalls the server and CLI globally, short alias ocu. On macOS run once and grant Accessibility and Screen Recording.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Drives the real desktop session: clicks, types and can change data in apps
- Requires Accessibility and Screen Recording permissions
This is third-party code. Review the repository files before installing.
What it does
The project wraps computer control in a local MCP server and CLI that any agent can connect to. Through the operating system accessibility layer it captures a semantic tree of an app's interface and acts on elements rather than only on coordinates. The tool surface is the same on every platform: list apps, get app state, click, secondary action, scroll, drag, type text, press key and set value. It ships install commands for Codex, Claude Code, Gemini CLI and opencode, plus a manual JSON config fragment. On macOS it needs version 14.0 or later and Accessibility and Screen Recording permissions.
Who it is for. For developers and QA engineers who need an agent to drive desktop apps directly.
Good fit when
- You need the agent to click and type in ordinary desktop apps
- You need computer use locally, without a cloud provider
- You need one control tool surface across macOS, Linux and Windows
Not a fit when
- The task is solved by browser automation rather than full desktop control
- macOS is older than 14.0, where the runtime does not launch
- You cannot give the agent access to a live desktop session
Example request
Open TextEdit, create a new document and type my note into itMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
OPEN_COMPUTER_USE_ALLOW_GLOBAL_POINTER_FALLBACKS- Enables global pointer fallbacks on macOS, including window-server drag sessions. It may move the real pointer, so enable it only when needed.
Limitations
The server drives the user's real desktop session, so it acts like a person at the machine. macOS needs version 14.0 or later plus Accessibility and Screen Recording permissions. On macOS a drag does not run a window-server drag session unless OPEN_COMPUTER_USE_ALLOW_GLOBAL_POINTER_FALLBACKS is enabled. A model for the agent must be connected separately; it is not part of the project.
How to disable. Remove the open-computer-use block from your agent's MCP config and uninstall the global package: npm rm -g open-computer-use.
FAQ
Which systems are supported?
macOS 14.0 and later, Linux and Windows. Each platform has its own runtime, but the tool surface is identical.
Is a cloud service required?
No. The server is local and runs on your machine. You connect the agent's model provider yourself.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that lets an agent see the screen and control macOS, Windows and Linux: screenshots, clicks, OCR and screen memory
computer-use-linux is an MCP server with an on-screen agent activity indicator that lets user clicks reach the application.
A Swift MCP server that lets an agent drive any macOS app: click, type and read the interface tree
MCP server for macOS desktop control through the Accessibility API: clicks, typing, windows, apps, shell commands and AppleScript
An MCP server for macOS: the agent moves the mouse, types, manages windows and captures the screen when an app has no API
An MCP server for driving Windows: the agent opens apps, clicks and types, reads window and UI state, and works with files and PowerShell
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Open Computer Use (this card) | no | not required | 2,940 | |
| agent-desktop | no | 22,602 | a Rust CLI for macOS | |
| Ghost OS | no | not required | 1,686 | a Swift MCP server for macOS |
| mcp-server-macos-use | no | not required | 357 | a Swift MCP server that lets an agent drive any macOS app |
README in short
The README presents open-computer-use as an open alternative to Codex Computer Use wrapped as MCP. It installs globally via npm, with ocu as a short command alias. It offers install commands for Codex, Claude Code, Gemini CLI and opencode, a manual JSON config fragment and a skill install via npx skills. It shows examples of calling individual tools and sequences through the CLI, plus a doctor command to check permissions. MIT licensed.