This project provides a Windows server for the Model Context Protocol (MCP), enabling AI models to programmatically control system operations like mouse, keyboard, window management, and screen capture. It targets AI researchers and developers seeking to integrate AI with desktop automation, offering secure, structured control for tasks ranging from testing to complex human-AI collaboration.

How It Works

This project bridges AI models and the Windows desktop via MCP. It exposes core system functionalities through a server architecture, allowing AI agents to send commands for precise mouse, keyboard, window, and screen interactions. MCPControl leverages pluggable automation providers (keysender, powershell, autohotkey) for underlying OS operations, offering flexibility and native Windows integration. Its design prioritizes secure, programmatic access, supporting SSE and HTTPS transports.

Quick Start & Requirements

• Installation: npm install -g mcp-control
• Prerequisites: Windows OS, Node.js (LTS), npm, Python 3.12, and Visual Studio Build Tools (with VC++ workload). Native build tools (node-gyp, cmake-js) are needed for development.
• Configuration: Optimal performance is achieved in a VM at 1280x720 resolution. The Claude client must connect via SSE transport, specifying the server's IP/port or a local launch command.
• Running the Server: Execute mcp-control --sse.
• Docs: Model Context Protocol Documentation (URL not provided).

Highlighted Details

• Flexible Automation: Supports keysender (default), powershell, and autohotkey (v2) providers.
• Security: Mandates HTTPS/TLS for production remote deployments.
• Use Cases: AI-driven application testing, workflow automation, AI gaming, visual reasoning, and cross-application integration.
• Comprehensive Control: Features for mouse, keyboard, window management, and screen capture.

Maintenance & Community

No specific maintainer or community channel details were found. Contribution guidelines are available via CONTRIBUTING.md.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Strictly Windows-only. Optimized for 1280x720 resolution on a single screen.

Limitations & Caveats

This software is experimental and risky, granting AI direct system control; users assume all responsibility. Window minimize/restore operations are unsupported. Multi-screen functionality may be unreliable, and get_screenshot has issues with VS Code Extension Cline. Some operations may require elevated permissions. Click accuracy is resolution-dependent, with ongoing bug fixes.