Munim Computer Use — Claude Code MCP server for desktop control
An MCP server that lets a coding agent read the screen and drive apps: background clicks and typing while your mouse stays free.
claude mcp add --env COMPUTER_USE_BROWSER=<COMPUTER_USE_BROWSER> --env COMPUTER_USE_AGENT_CURSOR=<COMPUTER_USE_AGENT_CURSOR> --transport stdio munim-computer-use -- npx -y munim-computer-useWhat Munim Computer Use does
The agent gets the accessibility tree of the open window, so buttons and fields arrive as elements with ids rather than pixels. It starts with get_app_state, acts on an element id, then checks the outcome with screenshot. No vision model is needed for interaction.
macOS, Windows and Linux are supported. Input happens in the background, which leaves your own cursor usable while the agent works. There is also an agent pointer showing where it is acting, plus zoom for small text. The server can drive tabs in your signed-in Chrome. Two settings, COMPUTER_USE_BROWSER and COMPUTER_USE_AGENT_CURSOR, are passed as environment variables when you register the server.
On macOS you must grant Accessibility and Screen Recording first; run munim-computer-use request-permissions once by hand. Without those permissions the agent can neither see the screen nor press anything. Windows and Linux skip that step.
What it helps with
- read the state of a program window through its accessibility tree
- click and type without taking the mouse away from the user
- magnify a small piece of text before acting on it
- work through tabs in a Chrome session that is already signed in
- take a screenshot to confirm a step worked
Example prompts for the agent
- Open Safari, find the cheapest Tuesday flight to Denver and put it in a note
- Read the current window with
get_app_stateand press the Export button - Switch to the order tab in Chrome and check the delivery status
How to install Munim Computer Use
Platform CLI
claude mcp add --env COMPUTER_USE_BROWSER=<COMPUTER_USE_BROWSER> --env COMPUTER_USE_AGENT_CURSOR=<COMPUTER_USE_AGENT_CURSOR> --transport stdio munim-computer-use -- npx -y munim-computer-useReplace <…> with your own values
Platform CLI
codex mcp add munim-computer-use --env COMPUTER_USE_BROWSER=<COMPUTER_USE_BROWSER> --env COMPUTER_USE_AGENT_CURSOR=<COMPUTER_USE_AGENT_CURSOR> -- npx -y munim-computer-useReplace <…> with your own values
Config file
{
"mcpServers": {
"munim-computer-use": {
"command": "npx",
"args": [
"-y",
"munim-computer-use"
],
"env": {
"COMPUTER_USE_BROWSER": "<COMPUTER_USE_BROWSER>",
"COMPUTER_USE_AGENT_CURSOR": "<COMPUTER_USE_AGENT_CURSOR>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"munim-computer-use": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"munim-computer-use"
],
"env": {
"COMPUTER_USE_BROWSER": "<COMPUTER_USE_BROWSER>",
"COMPUTER_USE_AGENT_CURSOR": "<COMPUTER_USE_AGENT_CURSOR>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e COMPUTER_USE_BROWSER=<COMPUTER_USE_BROWSER> -e COMPUTER_USE_AGENT_CURSOR=<COMPUTER_USE_AGENT_CURSOR> munim-computer-use npx -y munim-computer-useReplace <…> with your own values
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 Munim Computer Use
How do I install Munim Computer Use in Claude Code?
Run claude mcp add munim-computer-use -- npx -y munim-computer-use. For the browser and pointer flags, use the other form: `claude mcp add --env COMPUTER_USE_BROWSER=<COMPUTER_USE_BROWSER> --env COMPUTER_USE_AGENT_CURSOR=<COMPUTER_USE_AGENT_CURSOR> --transport stdio munim-computer-use -- npx -y munim-computer-use`.
What does Munim Computer Use do?
It gives an agent desktop access: interface reading via the accessibility tree, background clicks and typing, screenshots, and control of Chrome tabs.
What does it need on macOS?
Run munim-computer-use request-permissions once and approve Accessibility and Screen Recording, otherwise the agent cannot see the screen.
