CursorTouch/Windows-MCP

What it solves

Windows-MCP provides a bridge between Large Language Models (LLMs) and the Windows operating system. It allows AI agents to interact with a computer as a human would—navigating files, controlling applications, and interacting with the UI—without requiring specialized computer vision models or fine-tuned AI for every task.

How it works

It operates as a Model Context Protocol (MCP) server that exposes a set of tools to an MCP-compatible client (like Claude Desktop). The server interacts natively with Windows UI elements and simulates user inputs (keyboard and mouse). It also includes a specialized "DOM Mode" for browser automation that filters out browser UI elements to focus specifically on web page content for Chrome, Edge, and Firefox.

Who it’s for

This tool is designed for developers and AI researchers building AI agents that need to automate Windows tasks, perform QA testing, or create system-level integrations for LLMs.

Highlights

• Native Windows Integration: Controls windows, opens apps, and simulates user input across Windows 7 through 11.
• LLM Agnostic: Works with any LLM without needing traditional computer vision or specific fine-tuning.
• Browser Automation: Dedicated DOM mode for cleaner web automation in major browsers.
• Low Latency: Action latency typically ranges between 0.2 to 0.5 seconds.
• Flexible Deployment: Supports multiple transport methods including stdio, SSE, and streamable-HTTP, with built-in security features like OAuth 2.0, TLS, and IP allowlisting.