mcp-windows-desktop-automation

MCP.Pizza Chef: mario-andreschak

Fifty tools drive a Windows PC directly: moving and clicking the mouse, typing keystrokes, reading and setting the clipboard, finding, resizing and closing windows, filling in form fields, launching programs, and even shutting the machine down. Windows only, and you build it from source yourself. Nothing is sandboxed and nothing asks before it acts. The screenshot feature the description advertises was never written — it returns a placeholder sentence instead of a picture. The code has not changed since March 2025.

Unmaintained · No commits in 18 months.
Other

Use This MCP server To

Fill in the same desktop form without touching the keyboard Close the pile of windows cluttering up my screen Launch a program and wait for it to finish Read out whatever is currently on my clipboard Resize and line up windows the way I like them Click through a repetitive installer step by step

README

MCP Windows Desktop Automation

A Model Context Protocol (MCP) server for Windows desktop automation using AutoIt.

Overview

This project provides a TypeScript MCP server that wraps the node-autoit-koffi package, allowing LLM applications to automate Windows desktop tasks through the MCP protocol.

The server exposes:

  • Tools: All AutoIt functions as MCP tools
  • Resources: File access and screenshot capabilities
  • Prompts: Templates for common automation tasks

Features

  • Full wrapping of all AutoIt functions as MCP tools
  • Support for both stdio and WebSocket transports
  • File access resources for reading files and directories
  • Screenshot resources for capturing the screen or specific windows
  • Prompt templates for common automation tasks
  • Strict TypeScript typing throughout

Installation

# Clone the repository
git clone https://github.com/yourusername/mcp-windows-desktop-automation.git
cd mcp-windows-desktop-automation

# Install dependencies
npm install

# Build the project
npm run build

Usage

Starting the Server

# Start with stdio transport (default)
npm start

# Start with WebSocket transport
npm start -- --transport=websocket --port=3000

# Enable verbose logging
npm start -- --verbose

mcp-windows-desktop-automation FAQ

Does it work on a Mac?
No. It is Windows only, because it drives the desktop through AutoIt, which does not exist on macOS or Linux.
Do I need a key or an account?
No. There is no account and no key — but that also means anything connected to it can act as you on your own PC.
Can it take screenshots like the description says?
No. That part was never finished. Asking for a screenshot returns the placeholder text 'Screenshot of full screen would be captured here'.
Is anything protected from mistakes?
No. It can launch any program, send any keystroke and shut the computer down, with no confirmation prompt and no sandbox.
Should I give it my Windows password?
One tool asks for your Windows username and password so it can run a program as another user, which puts that password into the conversation. Avoid it.
How hard is the setup?
Hard. You download the code, install its parts and build it yourself, and the download link in the instructions is still an unfilled placeholder.
Is it still maintained?
No. Nothing has changed since March 2025, and several advertised pieces were left unfinished. It is MIT licensed, so workplace use is permitted.
Does the remote connection option work?
No. Choosing it prints 'WebSocket transport not fully implemented yet' and quits, so only the standard local connection works.