computer-control
MCP server that provides computer control capabilities, like mouse, keyboard, OCR, etc. using PyAutoGUI, RapidOCR, ONNXRuntime. Similar to 'computer-use' by Anthropic. With Zero External Dependencies.
claude mcp add --transport stdio ab498-computer-control-mcp uvx computer-control-mcp@latest \ --env COMPUTER_CONTROL_MCP_WGC_PATTERNS="Optional: comma-separated window title patterns to force Windows Graphics Capture (WGC) for matching windows" \ --env COMPUTER_CONTROL_MCP_SCREENSHOT_DIR="Optional: directory to save screenshots (defaults to downloads directory if not set)"
How to use
This MCP server provides a Python-based computer control interface. It exposes tools to manipulate the mouse, simulate keyboard input, capture screenshots (with optional OCR), and manage windows. The server runs via uvx (Python) and downloads the underlying Python package computer-control-mcp on first run. Once started, you can issue commands through the API exposed by the package to perform actions such as moving or clicking the mouse, typing text, taking screenshots (with optional OCR extraction), listing and activating windows, and performing drag-and-drop operations. The feature set is designed to resemble automated UI control utilities while keeping external dependencies minimal.
To use, install via uvx computer-control-mcp@latest (or install with pip and run the executable name), then interact with the available tools through the provided API. You’ll typically start the server in your environment and call methods like take_screenshot, type_text, click_screen, and activate_window to automate GUI tasks. The server also supports configuration options via environment variables (for screenshot storage and WGC behavior) to tailor behavior to your OS and windowing environment.
How to install
Prerequisites:
- Python installed (recommended 3.8+)
- uv (uvx) package manager or pip
- Basic access to install Python packages from PyPI
Option A — Install and run via uvx (recommended):
- Install or upgrade uvx: pip install uvx --upgrade
- Start the MCP server (this will download computer-control-mcp on first run): uvx computer-control-mcp@latest
Option B — Install in development and run locally:
- Ensure Python is installed
- Install the package in editable mode: pip install -e .
- Run the server (equivalent to uvx computer-control-mcp@latest): computer-control-mcp
Notes:
- If you want to force usage of the latest local code, use the development route rather than the uvx alias.
- For Windows, you may need additional permissions or firewall rules to allow GUI automation.
Testing:
- You can run tests with: python -m pytest
Additional notes
Tips and common questions:
- The environment variable COMPUTER_CONTROL_MCP_SCREENSHOT_DIR lets you redirect where screenshots are saved. If the directory does not exist, the server will fall back to the default downloads location.
- To avoid black screens on GPU-accelerated Windows apps, enable Windows Graphics Capture (WGC) for specific windows by setting COMPUTER_CONTROL_MCP_WGC_PATTERNS to a comma-separated list of patterns that match window titles (e.g., obs, discord, game, steam).
- If you’re integrating with other automation tools, you can query and control windows using list_windows and activate_window, enabling automation workflows that require foreground focus.
- The note on first-run download time applies to the initial setup; subsequent runs will be faster as dependencies are cached.
- This MCP is Python-based and uses PyAutoGUI and OCR tooling under the hood; ensure you have the necessary system permissions for screen capture and input simulation on your platform.
Related MCP Servers
gpt-researcher
An autonomous agent that conducts deep research on any data using any LLM providers.
cuga-agent
CUGA is an open-source generalist agent for the enterprise, supporting complex task execution on web and APIs, OpenAPI/MCP integrations, composable architecture, reasoning modes, and policy-aware features.
anything-to-notebooklm
Claude Skill: Multi-source content processor for NotebookLM. Supports WeChat articles, web pages, YouTube, PDF, Markdown, search queries → Podcast/PPT/MindMap/Quiz etc.
example
A ready-to-use MCP (Model Context Protocol) server template for extending Cursor IDE with custom tools. Deploy your own server to Heroku with one click, create custom commands, and enhance your Cursor IDE experience. Perfect for developers who want to add their own tools and commands to Cursor IDE without complex setup.
hyperliquid
🚀 MCP Server para Hyperliquid DEX - Trade com Claude usando linguagem natural. Desenvolvido por Caio Vicentino para as comunidades Yield Hacker, Renda Cripto e Cultura Builder
mcp-ssh-orchestrator
Secure SSH access for AI agents via MCP. Execute commands across your server fleet with policy enforcement, network controls, and comprehensive audit logging.