An MCP server that provides computer control capabilities, including mouse and keyboard control, OCR, and screen management, using PyAutoGUI, RapidOCR, ONNXRuntime without external dependencies.

Installation

UVX

uvx computer-control-mcp@latest

Global Installation via Pip

pip install computer-control-mcp
computer-control-mcp

From Source Code

git clone https://github.com/AB498/computer-control-mcp.git
cd computer-control-mcp
pip install -e .
python -m computer_control_mcp.core

Build with Hatch

pip install hatch
hatch build
pip install dist/*.whl --upgrade
computer-control-mcp

Configuration

Claude Desktop

{
  "mcpServers": {
    "computer-control-mcp": {
      "command": "uvx",
      "args": ["computer-control-mcp@latest"]
    }
  }
}

Available Tools

Tool	Description
click_screen	Click at specified screen coordinates
move_mouse	Move mouse cursor to specified coordinates
drag_mouse	Drag mouse from one position to another
mouse_down	Hold down a mouse button ('left', 'right', 'middle')
mouse_up	Release a mouse button ('left', 'right', 'middle')
type_text	Type specified text at current cursor position
press_key	Press specified key
key_down	Hold down a specific key until release
key_up	Release a specific key
press_keys	Press keys (supports individual keys, sequences, and combinations)
take_screenshot	Capture screen or window with optional save to downloads folder
take_screenshot_with_ocr	Extract and return text with coordinates via OCR from screen or window
get_screen_size	Get current screen resolution
list_windows	List all open windows
activate_window	Bring specified window to foreground

Features

• Mouse movement and click control
• Text input at current cursor position
• Full screen or specific window screenshots with optional save to downloads folder
• Text extraction from screenshots using OCR (Optical Character Recognition)
• Window viewing and activation
• Key pressing
• Drag operations

Notes

Running uvx computer-control-mcp@latest for the first time will download Python dependencies (approximately 70MB), which may take some time. It is recommended to run this in a terminal before using as an MCP. Subsequent runs will be instantaneous.