MCPControl

What is MCPControl?

MCPControl is a Windows-only control server for the Model Context Protocol (MCP) that provides programmatic access to mouse, keyboard, window management, screen capture, and clipboard operations. It enables AI models like Claude to securely interact with desktop applications.

How to use MCPControl?

Install globally via npm install -g mcp-control. Configure your MCP client (e.g., Claude) with SSE transport – either by pointing to a running server’s URL (e.g., http://192.168.1.100:3232/mcp) or by launching MCPControl locally with --sse. Start the server with mcp-control --sse. For best click accuracy, run inside a virtual machine at 1280×720 resolution.

Key features of MCPControl

• 🖱️ Precision mouse movement, clicks, drags, and scrolling
• ⌨️ Keyboard text input, key combos, press/release, and hold
• 🪟 Window listing, focusing, resizing, and repositioning
• 📸 High-quality screenshots and screen size detection
• 📋 Clipboard read/write operations
• 🔧 Multiple automation providers (keysender, PowerShell, AutoHotkey)

Use cases of MCPControl

• Assisted Automation: Delegate UI testing, workflow automation, and form filling to AI
• AI Experimentation: Let Claude learn to play simple games or solve visual puzzles
• Development and Testing: Create cross-application integration, UI testing, and automated demo creation

FAQ from MCPControl

What platforms does MCPControl support?

Only Windows is supported. The project is not compatible with macOS or Linux.

Do I need HTTPS for remote access?

Yes, according to the MCP specification, HTTPS is mandatory for all HTTP-based transports in production. Use the --https flag with valid TLS certificates when deploying remotely.

What automation providers are available?

Three providers: keysender (default, native Windows automation), powershell (simpler operations), and autohotkey (advanced scripting with AutoHotkey v2). Providers can be mixed per operation via environment variables.

Are there any known limitations?

Window minimize/restore is unsupported. Multi-screen functions may not work reliably. The get_screenshot tool does not work with the VS Code extension Cline. Click accuracy is optimized for 1280×720 single-screen setups.

What dependencies are required to build from source?

Node.js 18+, npm, and native build tools (node-gyp, cmake-js) are needed. Windows SDK with VC++ workload must be installed for the keysender native module.