Windows-MCP

Windows-MCP is a lightweight, open-source server that lets AI agents interact with the Windows operating system. It bridges LLMs with Windows tasks such as navigating files, controlling applications, interacting with the UI, and performing QA testing, enabling automation and testing workflows across the desktop.

How to use

You connect a compatible MCP client to Windows-MCP to start automating Windows tasks. Use the client to issue high-level actions like opening applications, navigating the file system, clicking UI elements, typing into fields, or reading window states. The server exposes a set of tools that allow you to perform keyboard and mouse input, window management, and state capture in real time. Typical workflows include launching an app, locating a file, performing a sequence of UI actions, and validating results through screenshots or state snapshots.

How to install

Prerequisites you need before running Windows-MCP are:
- Python 3.13 or newer
- UV (Desktop Extension package manager) installed with "pip install uv"
- Claude Desktop or another MCP client installed
- English language set as the default in Windows (or configure the server to run with a different language by disabling certain tools)

To enable Windows-MCP as an MCP server in your client, follow this concrete setup using the Gemini configuration path provided by your client:

• Open your client’s settings file located at the user profile path and add the Windows-MCP configuration under MCP servers.

The dedicated MCP server configuration uses a local runtime with UV to run the main Python script. You would wire it in your client’s settings as shown below. This enables the client to start the Windows-MCP process directly on your machine.

Configuration example

{
  "theme": "Default",
  ...
  //MCP Server Config
  "mcpServers": {
    "windows-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "<path to the windows-mcp directory>",
        "run",
        "main.py"
      ]
    }
  }
}

Security and notes

This MCP interacts directly with Windows to perform actions. Use it only in trusted environments and with explicit permission to automate tasks on the target machine.

Troubleshooting and tips

If you encounter issues, verify that the UV runtime can access the Windows-MCP directory and that the Python script path is correct. Check client logs for startup errors, and ensure the MCP client has the necessary permissions to simulate input and access UI elements.

Available tools

Click-Tool

Clicks on a screen position using absolute coordinates to simulate mouse clicks.

Type-Tool

Types text into a focused element, optionally clearing existing content.

Clipboard-Tool

Copies to or pastes from the system clipboard.

Scroll-Tool

Scrolls vertically or horizontally within a window or region.

Drag-Tool

Drags the mouse from one coordinate to another.

Move-Tool

Moves the mouse pointer to a specific position.

Shortcut-Tool

Sends keyboard shortcuts like Ctrl+C or Alt+Tab.

Key-Tool

Presses a single key.

Wait-Tool

Pauses actions for a defined duration.

State-Tool

Captures a snapshot of language, active apps, UI state, and a desktop screenshot.

Resize-Tool

Resizes a window or moves it to a new location.

Launch-Tool

Launches an application from the start menu.

Shell-Tool

Executes PowerShell commands.

Scrape-Tool

Scrapes information from the current webpage or UI region.