macOS Control MCP Server
Enables full desktop automation on macOS through natural language, including mouse control, keyboard input, screen capture, and GUI interaction using PyAutoGUI.
README
macOS Control MCP Server
A Model Context Protocol (MCP) server for macOS desktop automation using FastMCP and PyAutoGUI. This MCP gives full control to your MacOS desktop. It will work with Linux, and windows but these are not tested. Needless to day this is very dangerous. Run in a container, VM, etc.
Features
Mouse Control
- Move mouse to coordinates
- Click (left, right, middle)
- Double-click
- Scroll
- Drag and drop
- Get mouse position
Keyboard Control
- Type text
- Press individual keys
- Execute hotkey combinations (Cmd+C, Cmd+V, etc.)
- Hold and release keys
Screen Capture & Analysis
- Take screenshots (full screen or regions)
- Get screen resolution
- Locate images on screen
- Get pixel colors at coordinates
Utilities
- Sleep/pause execution
- Configure failsafe and pause settings
Installation
Prerequisites
- macOS
- Python 3.10 or later
- uv package manager
Setup
-
Clone or navigate to the project directory:
cd ~/macos-control-mcp -
Install dependencies:
uv sync -
Grant necessary permissions:
- Accessibility Access: System Settings → Privacy & Security → Accessibility
- Screen Recording: System Settings → Privacy & Security → Screen Recording
Add Terminal (or your terminal app) to both of these permissions.
Configure Claude Desktop For development
Add this to your Claude Desktop configuration file (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"macos-control": {
"command": "uv",
"args": [
"--directory",
"<path_to_folder>/macos-control-mcp",
"run",
"macos-control-mcp"
]
}
}
}
Usage Examples
Once configured in Claude Desktop, you can use natural language to control your Mac:
Mouse Control
- "Move the mouse to position 500, 300"
- "Click at coordinates 100, 200"
- "Double-click at the center of the screen"
- "Right-click at 400, 500"
- "Scroll down 5 clicks"
Keyboard Control
- "Type 'Hello World'"
- "Press the Enter key"
- "Press Cmd+C to copy"
- "Press Cmd+Shift+3 to take a screenshot"
Screen Capture
- "Take a screenshot"
- "What's the screen resolution?"
- "Get the color of the pixel at 100, 100"
Combined Actions
- "Move mouse to 500, 300 and click"
- "Type 'test@example.com' and press Tab"
- "Press Cmd+Space, type 'Safari', and press Enter"
Available Tools
Mouse Tools
move_mouse(x, y, duration)- Move cursor to coordinatesclick_mouse(x, y, button, clicks, interval)- Click at positiondouble_click(x, y)- Double-clickright_click(x, y)- Right-clickscroll_mouse(clicks, x, y)- Scroll wheeldrag_mouse(x, y, duration, button)- Drag to positionget_mouse_position()- Get current cursor position
Keyboard Tools
type_text(text, interval)- Type textpress_key(key, presses, interval)- Press a keyhotkey(*keys)- Press key combinationkey_down(key)- Hold down a keykey_up(key)- Release a key
Screen Tools
get_screen_size()- Get screen resolutiontake_screenshot(region)- Capture screenlocate_on_screen(image_path, confidence)- Find imageget_pixel_color(x, y)- Get RGB color
Utility Tools
sleep(seconds)- Pause executionset_failsafe(enabled)- Configure failsafeset_pause(duration)- Set action pause
Safety Features
Failsafe
By default, moving the mouse to any screen corner will abort PyAutoGUI operations. This can be disabled with set_failsafe(False).
Pause
A 0.1 second pause is inserted between actions by default. Adjust with set_pause(duration).
Development
Running Tests
uv run pytest
Running the Server Manually
uv run macos-control-mcp
Troubleshooting
"Permission denied" errors
Make sure you've granted Accessibility and Screen Recording permissions to your terminal application.
"Image not found" errors
When using locate_on_screen, try lowering the confidence parameter (default is 0.9).
Server not responding
- Check Claude Desktop logs:
~/Library/Logs/Claude/ - Verify the path in your configuration matches your installation
- Restart Claude Desktop
License
MIT
Credits
Built with:
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。