Windows

Windows-MCP is a lightweight MCP server that bridges AI agents with the Windows operating system. It provides a set of tools for interacting with Windows UI elements, launching and controlling applications, navigating the filesystem, and performing UI automation tasks. The server is designed to work with any large language model (LLM) without relying on traditional computer vision pipelines, and it includes a DOM mode (use_dom=True) for focused browser automation when needed. You can use it directly from PyPI via uvx (recommended) or run from source with uv.

To use it, install the MCP server and run it through your preferred integration (Claude Desktop, Perplexity Desktop, Gemini CLI, or any environment that supports MCP configuration). The server exposes a Windows-automation surface that your agent can call through the MCP protocol, enabling actions such as opening apps, navigating windows, simulating keyboard and mouse input, capturing UI state, and performing QA tests. For web automation inside a browser, you can enable DOM mode to filter browser chrome and focus on page content for cleaner automation results.

Prerequisites

• Python 3.13+ installed on Windows
• UV (Package Manager) from Astra, install with either:
  • pip: pip install uv
  • or run the installer script: curl -LsSf https://astral.sh/uv/install.sh | sh

Installation options A) Install from PyPI (recommended)

1. Install UV and the Windows-MCP package from PyPI: uvx windows-mcp
2. Verify installation by listing the command: uvx --version

B) Install from source

1. Clone the repository: git clone https://github.com/CursorTouch/Windows-MCP.git cd Windows-MCP
2. Run the server from source via UV: uv --directory . run windows-mcp

Configuration for Claude Desktop / Perplexity / Gemini

• Claude Desktop / Perplexity / Gemini can load the MCP server configuration directly. Use the following example configuration to connect: { "mcpServers": { "windows-mcp": { "command": "uvx", "args": [ "windows-mcp" ] } } }

First run notes

• The initial installation may take a minute or two to install dependencies listed in pyproject.toml. On the first run the server may timeout; if that happens, restart the server and let it complete its setup.
• Ensure the system language is English by default, or disable the App-Tool in the MCP Server if using a non-English Windows installation.

Tips and common issues:

• Environment: Windows 7 through Windows 11 are supported. Ensure Python 3.13+ and UV are installed.
• If using Claude Desktop or Perplexity Desktop, consider using the PyPI install (uvx) for a smoother experience. When using the source install, you may need to provide a full absolute path to uv.exe on systems where PATH does not include its location.
• If your MCP server is not discovered by the editor, verify your JSON config uses the correct command and path to the windows-mcp entry point. In some environments, using the full path to uv.exe helps: "C:\Users<user>.local\bin\uv.exe".
• Use the DOM mode (use_dom=True) for browser automation to reduce interference from browser UI elements.
• The Windows-MCP project is designed to be lightweight with minimal dependencies; if you encounter dependency installation issues, ensure you have a working Python environment and network access to fetch packages from PyPI.
• For Claude Desktop integration, review the example config files and ensure UTF-8 encoding without BOM to avoid JSON parsing issues.