omni-computer-use — Windows MCP server
Hands an agent the Windows desktop: screenshots, clicks, keystrokes. Useful where built-in computer use is missing.
claude mcp add --env COMPUTER_USE_MAX_PIXELS=<COMPUTER_USE_MAX_PIXELS> --env COMPUTER_USE_MASKING=<COMPUTER_USE_MASKING> --env COMPUTER_USE_ENFORCE_FOREGROUND=<COMPUTER_USE_ENFORCE_FOREGROUND> --env COMPUTER_USE_DEV=<COMPUTER_USE_DEV> --transport stdio omni-computer-use -- uvx omni-computer-use-mcpWhat omni-computer-use does
The server mirrors the computer-use tool surface from Claude Desktop, but ships as a plain MCP server with no client integration. It carries 29 tools: 27 line up with Anthropic's set, plus deactivate and display_overview. Setting COMPUTER_USE_DEV=on adds a reload tool, bringing the count to 30.
Click, move, scroll and drag coordinates live in the pixel space of the latest screenshot; the server converts them to physical pixels. On a multi-monitor machine each capture names the display it came from and lists the others. Synthetic keystrokes are refused when the agent's own window sits in front — otherwise the text would land in its conversation or run as a shell command.
Windows and Python 3.11+ are required, and uv/uvx fetch Python for you. The PyPI package is omni-computer-use-mcp while the command, server key and repo stay omni-computer-use. Installing into Claude once exposes the server to both the CLI and the Claude Desktop app, since Desktop reads Claude's MCP servers alongside its own.
What it helps with
- launch an app and type into its window
- capture the right monitor on a multi-display setup
- send a click or scroll to a specific screen point
- read from and write to the clipboard
- let an agent debug your own desktop app
Example prompts for the agent
- Take a screenshot and tell me which monitor it is
- Open Calculator and work out 12 times 8
- Switch to the second display and describe what is open there
How to install omni-computer-use
Platform CLI
claude mcp add --env COMPUTER_USE_MAX_PIXELS=<COMPUTER_USE_MAX_PIXELS> --env COMPUTER_USE_MASKING=<COMPUTER_USE_MASKING> --env COMPUTER_USE_ENFORCE_FOREGROUND=<COMPUTER_USE_ENFORCE_FOREGROUND> --env COMPUTER_USE_DEV=<COMPUTER_USE_DEV> --transport stdio omni-computer-use -- uvx omni-computer-use-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from omni-computer-use-mcp <command>
Platform CLI
codex mcp add omni-computer-use --env COMPUTER_USE_MAX_PIXELS=<COMPUTER_USE_MAX_PIXELS> --env COMPUTER_USE_MASKING=<COMPUTER_USE_MASKING> --env COMPUTER_USE_ENFORCE_FOREGROUND=<COMPUTER_USE_ENFORCE_FOREGROUND> --env COMPUTER_USE_DEV=<COMPUTER_USE_DEV> -- uvx omni-computer-use-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from omni-computer-use-mcp <command>
Config file
{
"mcpServers": {
"omni-computer-use": {
"command": "uvx",
"args": [
"omni-computer-use-mcp"
],
"env": {
"COMPUTER_USE_MAX_PIXELS": "<COMPUTER_USE_MAX_PIXELS>",
"COMPUTER_USE_MASKING": "<COMPUTER_USE_MASKING>",
"COMPUTER_USE_ENFORCE_FOREGROUND": "<COMPUTER_USE_ENFORCE_FOREGROUND>",
"COMPUTER_USE_DEV": "<COMPUTER_USE_DEV>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from omni-computer-use-mcp <command>
Config file
{
"servers": {
"omni-computer-use": {
"type": "stdio",
"command": "uvx",
"args": [
"omni-computer-use-mcp"
],
"env": {
"COMPUTER_USE_MAX_PIXELS": "<COMPUTER_USE_MAX_PIXELS>",
"COMPUTER_USE_MASKING": "<COMPUTER_USE_MASKING>",
"COMPUTER_USE_ENFORCE_FOREGROUND": "<COMPUTER_USE_ENFORCE_FOREGROUND>",
"COMPUTER_USE_DEV": "<COMPUTER_USE_DEV>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from omni-computer-use-mcp <command>
Platform CLI
gemini mcp add -e COMPUTER_USE_MAX_PIXELS=<COMPUTER_USE_MAX_PIXELS> -e COMPUTER_USE_MASKING=<COMPUTER_USE_MASKING> -e COMPUTER_USE_ENFORCE_FOREGROUND=<COMPUTER_USE_ENFORCE_FOREGROUND> -e COMPUTER_USE_DEV=<COMPUTER_USE_DEV> omni-computer-use uvx omni-computer-use-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from omni-computer-use-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about omni-computer-use
How do I install omni-computer-use in Claude Code?
Run claude mcp add omni-computer-use -- uvx omni-computer-use-mcp, then check claude mcp list — you want omni-computer-use … ✓ Connected in the output.
What does omni-computer-use do?
It gives an MCP client desktop control on Windows: screenshots, zoom, clicks, dragging, keyboard input, clipboard access and application launching.
Why do my keyboard actions get blocked?
type and key are refused whenever the agent's own window or its hosting terminal holds focus. That keeps the text out of the agent's conversation and stops it running as a shell command.
