Windows Computer Use MCP Server

English | 中文

An MCP (Model Context Protocol) server that provides desktop automation tools for Windows. Control mouse, keyboard, take screenshots, manage clipboard, and interact with applications — all through the MCP protocol.

Note: This server only runs on Windows.

Features

• Screenshots — Capture full screen or specific regions as base64 PNG

• Mouse Control — Move, click, double-click, right-click, drag, scroll

• Keyboard Input — Type text, press key combinations, hold keys

• Cursor Position — Get current cursor coordinates

• Clipboard — Read and write clipboard text

• Application Management — List running apps, get frontmost app, open apps

• Display Info — List connected displays with dimensions

Related MCP server: desktop-touch-mcp

Tools

Installation

Prerequisites

• Bun >= 1.3.11

• Windows 10/11

Install via npm

npm install -g windows-computer-use-mcp

Install via npx (no install needed)

npx windows-computer-use-mcp

Build from source

git clone https://github.com/hiphopanda/windows-computer-use-mcp.git
cd windows-computer-use-mcp
bun install
bun run build

Configuration

Add to your MCP client settings (e.g. Claude Desktop):

{
  "mcpServers": {
    "computer-use": {
      "type": "stdio",
      "command": "npx",
      "args": ["windows-computer-use-mcp"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "computer-use": {
      "type": "stdio",
      "command": "windows-computer-use-mcp"
    }
  }
}

Development

bun install          # Install dependencies
bun run dev          # Run in development mode
bun run build        # Build to dist/
bun run start        # Run built version

Architecture

• Runtime: Bun + TypeScript

• Transport: stdio (MCP protocol)

• Platform Layer: All desktop operations go through PowerShell scripts executed via Bun.spawn

• Low-level Input: Uses Win32 P/Invoke (user32.dll) loaded at runtime — no precompiled native modules needed

License

MIT

中文

一个基于 MCP（Model Context Protocol）协议的 Windows 桌面自动化服务器。通过 MCP 协议控制鼠标、键盘、截屏、剪贴板，以及管理应用程序。

注意：本服务器仅支持 Windows 系统。

功能特性

• 截屏 — 全屏或指定区域截图，返回 base64 PNG

• 鼠标控制 — 移动、单击、双击、右击、拖拽、滚轮滚动

• 键盘输入 — 输入文本、按键组合、长按按键

• 光标位置 — 获取当前光标坐标

• 剪贴板 — 读写剪贴板文本

• 应用管理 — 列出运行中的应用、获取前台应用、打开应用

• 显示器信息 — 列出已连接的显示器及分辨率

工具列表

安装

环境要求

• Bun >= 1.3.11

• Windows 10/11

通过 npm 安装

npm install -g windows-computer-use-mcp

通过 npx 运行（无需安装）

npx windows-computer-use-mcp

从源码构建

git clone https://github.com/hiphopanda/windows-computer-use-mcp.git
cd windows-computer-use-mcp
bun install
bun run build

配置

在 MCP 客户端配置中添加（如 Claude Desktop）：

{
  "mcpServers": {
    "computer-use": {
      "type": "stdio",
      "command": "npx",
      "args": ["windows-computer-use-mcp"]
    }
  }
}

或全局安装后：

{
  "mcpServers": {
    "computer-use": {
      "type": "stdio",
      "command": "windows-computer-use-mcp"
    }
  }
}

开发

bun install          # 安装依赖
bun run dev          # 开发模式运行
bun run build        # 构建到 dist/
bun run start        # 运行构建产物

架构

• 运行时：Bun + TypeScript

• 传输层：stdio（MCP 协议）

• 平台层：所有桌面操作通过 Bun.spawn 执行 PowerShell 脚本

• 底层输入：通过 Win32 P/Invoke（user32.dll）在运行时加载 — 无需预编译原生模块

许可证

MIT