Exposes Codex's computer-use runtime to any agent harness on macOS and Linux, without Codex authentication.
iFurySt/open-computer-use
An open-source Computer Use service exposed as an MCP server, letting AI agents control macOS, Linux, and Windows desktops.
Open-source alternative to:OpenAI Codex Computer Use
- Stars
- 2,512
- Forks
- 245
- Open issues
- 25
- Last commit
- yesterday
What is it?
open-computer-use is a Computer Use service wrapped as an MCP server. Any AI agent or MCP client can use it to observe and operate desktop applications on macOS, Linux, and Windows. It is positioned as an open-source alternative to OpenAI's Codex Computer Use and is designed to work without intruding on the desktop, building on the operating system's accessibility interfaces.
The project installs via npm and provides the open-computer-use command with a short ocu alias. It includes helper commands to register the server with Codex, Claude Code, Gemini CLI, and opencode, along with a CLI for calling individual tools, running tool sequences, and a JavaScript evaluation and REPL mode. On macOS, it requires version 14.0 or later and the Accessibility and Screen Recording permissions, and offers a doctor command to check them.
The repository also includes an agent skill that can be installed for Codex and Claude Code, plus a targeted query tool on macOS for finding controls by text or role without a screenshot.
Key features
- 01Exposes Computer Use actions as an MCP server for any MCP-compatible client.
- 02Runs on macOS, Linux, and Windows.
- 03Installs with a single npm command and offers an ocu alias.
- 04Provides one-step installers for Codex, Claude Code, Gemini CLI, and opencode.
- 05Includes a CLI for single tool calls, multi-step sequences, and JavaScript evaluation.
- 06Offers a permission check command on macOS.
Use cases
- Letting an AI coding agent such as Codex or Claude Code operate desktop apps during a task.
- Automating repetitive GUI workflows in applications that lack an API.
- Testing or driving desktop applications from scripts or agent pipelines.
Quick start
- 1
Install the package globally with npm
npm i -g open-computer-use - 2
On macOS, run once and grant Accessibility and Screen Recording permissions
open-computer-use - 3
Install into Codex (optional, or add the MCP JSON config manually)
open-computer-use install-codex-mcp
Things to know
- The macOS runtime requires macOS 14.0 or later.
- On macOS, Accessibility and Screen Recording permissions must be granted before use.
Topics
FAQ
- What is open-computer-use?
- An open-source Computer Use service exposed as an MCP server, letting AI agents control macOS, Linux, and Windows desktops. open-computer-use is a Computer Use service wrapped as an MCP server. Any AI agent or MCP client can use it to observe and operate desktop applications on macOS, Linux, and Windows. It is positioned as an open-source alternative to OpenAI's Codex Computer Use and is designed to work without intruding on the desktop, building on the operating system's accessibility interfaces.
- Is open-computer-use free to use? What license does it have?
- open-computer-use is open source under the MIT license. Check the license terms before using it in commercial products.
- Is open-computer-use still maintained?
- Actively maintained. The last commit to open-computer-use was on Oct 18, 2026.
- What can I use open-computer-use for?
- Letting an AI coding agent such as Codex or Claude Code operate desktop apps during a task. Automating repetitive GUI workflows in applications that lack an API. Testing or driving desktop applications from scripts or agent pipelines.
- How do I install open-computer-use?
- Install the package globally with npm: npm i -g open-computer-use On macOS, run once and grant Accessibility and Screen Recording permissions: open-computer-use Install into Codex (optional, or add the MCP JSON config manually): open-computer-use install-codex-mcp
- Where can I get open-computer-use?
- The source code and installation instructions are on GitHub: https://github.com/iFurySt/open-computer-use
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Lets AI agents like Claude Code or Codex read the screen, tap, type and swipe on a real iPhone over USB or Wi-Fi via MCP.
A macOS CLI and AI agent skill that draws big arrows, boxes and signs on screen to point users at UI elements, without clicking or capturing anything.
A local gateway and menu bar app that lets Claude Code, Codex, Gemini CLI and other coding agents use any model from any provider or subscription.
Long-term memory for AI coding agents that carries project context and handoffs across Claude Code, Codex, Cursor and other agent CLIs.
An agent skill that teaches AI coding assistants like Claude Code and Codex to write modern, correct SwiftUI code.