Desktop Controller MCP Server
Give any AI full control of your desktop. Mouse, keyboard, screenshots — all through the Model Context Protocol.
README
Desktop Controller MCP Server
Give any AI full control of your desktop. Mouse, keyboard, screenshots — all through the Model Context Protocol.
One MCP server. Ten tools. Your AI can now see your screen and interact with any application, just like you do.
What can it do?
- See your screen — Take full or partial screenshots, auto-optimized for LLM token usage
- Move & click — Navigate, click, double-click, right-click, drag and drop
- Type anything — Full keyboard control including Unicode, emoji, and key combos (Cmd+C, Ctrl+Shift+T, etc.)
- Cross-platform — Works on macOS, Windows, and Linux
Tools
| Tool | What it does |
|---|---|
desktop_screenshot |
Capture the screen or a region (returns image directly to the LLM) |
desktop_get_screen_size |
Get screen dimensions in logical pixels |
desktop_get_mouse_position |
Get current cursor coordinates |
desktop_move_mouse |
Move cursor (absolute or relative) |
desktop_click |
Click, double-click, or right-click at any position |
desktop_scroll |
Scroll up or down |
desktop_drag |
Drag from one position to another |
desktop_type_text |
Type text with full Unicode support |
desktop_press_key |
Press special keys (Enter, Tab, arrows, F1-F12, etc.) |
desktop_hotkey |
Press key combinations (Cmd+C, Ctrl+V, Alt+F4, etc.) |
Quick Start
Install
git clone https://github.com/YOUR_USERNAME/mcp-desktop-controller.git
cd mcp-desktop-controller
pip install -e .
Add to Claude Desktop
Add to your config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"desktop-controller": {
"command": "mcp-desktop-controller"
}
}
}
<details> <summary>Using a virtual environment? Use the full path instead.</summary>
{
"mcpServers": {
"desktop-controller": {
"command": "/path/to/venv/bin/mcp-desktop-controller"
}
}
}
</details>
Add to Claude Code
claude mcp add desktop-controller -- mcp-desktop-controller
macOS Permissions
On first run, you'll need to grant two permissions in System Settings > Privacy & Security:
- Accessibility — so the server can control mouse and keyboard
- Screen Recording — so it can take screenshots
Safety
- Failsafe — Move your mouse to the top-left corner (0,0) to instantly abort any action
- Action pause — 0.1s delay between all actions to keep things predictable
- Retina-aware — All coordinates use logical resolution; screenshots are auto-downscaled to save tokens
How it works
Built with FastMCP and PyAutoGUI. The server exposes desktop control as MCP tools over stdio. Any MCP-compatible client (Claude Desktop, Claude Code, or your own) connects and gets full desktop access.
Your AI ←→ MCP Client ←→ Desktop Controller ←→ Your Screen
Platform Support
| Platform | Status | Notes |
|---|---|---|
| macOS | Full support | Uses pbcopy for Unicode input |
| Windows | Full support | Uses clip for Unicode input |
| Linux | Full support | Requires xclip (sudo apt install xclip) |
Development
# Run directly
python server.py
# Test with MCP Inspector
fastmcp dev server.py
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。