desktop-automation-mcp
Windows desktop automation MCP server supporting mouse, keyboard, screenshot, and image recognition, with native CJK text input via clipboard.
README
desktop-automation-mcp
Windows desktop automation MCP server — screenshot, mouse, keyboard, and image recognition. Built for Claude Code and other MCP clients.
14 tools · Python 3.10+ · FastMCP · stdio transport
Features
- 🖱️ Mouse — click, double-click, right-click, move (smooth), drag
- ⌨️ Keyboard — type text, press single keys, key combinations (hotkeys)
- 📸 Screenshot — full screen or region capture
- 🔍 Image recognition — locate UI elements on screen by image template
- 🌐 CJK text input — Chinese/Japanese/Korean via clipboard paste
- 📏 Screen info — cursor position, screen resolution
- ⚠️ Alert — pop up desktop notification dialogs
Why this one?
Most desktop automation MCP servers are built by English-speaking developers and struggle with CJK text input. This server:
- Native CJK support — non-ASCII text automatically routes through clipboard paste
- Minimal & readable — 172 lines of Python, no black magic
- FastMCP stdio — latest MCP protocol, clean and reliable
Install
git clone https://github.com/Lianne0106/desktop-automation-mcp.git
cd desktop-automation-mcp
pip install -r requirements.txt
For locate_on_screen (image recognition), also install OpenCV:
pip install opencv-python
Claude Code setup
Add to ~/.claude/mcp.json:
{
"mcpServers": {
"desktop-automation": {
"command": "python",
"args": ["PATH\\TO\\desktop-automation-mcp\\server.py"],
"env": {
"PYTHONIOENCODING": "utf-8"
}
}
}
}
Restart Claude Code, then try:
> 截个屏看看
> 鼠标移到屏幕中央
> 帮我输入这段中文
Tools
| Tool | Description |
|---|---|
screenshot |
Capture full screen or region |
click |
Click at (x, y), any mouse button |
double_click |
Double-click at (x, y) |
right_click |
Right-click at (x, y) |
move |
Smoothly move cursor to (x, y) |
drag |
Click-drag from A to B |
type_text |
Type text (clipboard for CJK, keystrokes for ASCII) |
press_key |
Press a single key (enter, tab, escape, f1-f12, etc.) |
hotkey |
Key combination (ctrl+c, alt+tab, win+r, etc.) |
scroll |
Scroll wheel up/down, optionally at a position |
get_position |
Get current cursor position |
get_screen_size |
Get screen dimensions |
locate_on_screen |
Find an image template on screen |
alert |
Show a popup alert dialog |
Failsafe
Move mouse to the top-left corner to abort any running action (PyAutoGUI FAILSAFE).
Similar projects
- windows-computer-use-mcp — 22 tools, UI Automation, OCR, autonomous agent
- WindowsOSMCP — DPI-aware, multi-monitor, UIA element targeting
- pc-control-mcp — 30+ tools, Playwright browser, notes, clipboard
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。