Windows-MCP

Windows-MCP is an MCP server that gives your AI agents direct control over a Windows operating system.

It translates natural language commands from large language models into actual desktop interactions like clicking, typing, launching apps, and more.

You can use the MCP to automate software testing, perform repetitive UI tasks, or build complex agents that interact with desktop applications.

Key Features

• 🖱️ Desktop Control: Clicks, types, scrolls, and uses keyboard shortcuts.
• 📱 Application Management: Launches programs and manages windows.
• 🔍 State Inspection: Captures a snapshot of the desktop, including active apps and UI elements.
• 💻 Shell Access: Executes PowerShell commands directly.
• 🌐 Web Scraping: Extracts information from web pages.
• ⚡ Low Latency: Actions typically complete in 0.7 to 2.5 seconds.

USE CASES

• Automated QA Testing: Instruct an AI to test a new application build. It can open the software, navigate through menus, input test data, and verify expected behaviors without manual scripting.
• Routine Data Entry: Automate tedious tasks like transferring information from a spreadsheet into a legacy desktop application that lacks a modern API.
• Application Demo Agent: Build an AI that can demonstrate a complex software product live, responding to verbal requests like “show me how to create a new project” and performing the actions in real-time.
• IT Support Scripting: Create an agent that can help troubleshoot common user issues by guiding the AI to check specific system settings or restart services through the GUI.

How to Use It

1. To get started, you need Python 3.13+ and the uv package manager. It’s also recommended to have English as the default language on your Windows machine.

2. Integrate the MCP server with your AI assistangs.

For Claude Desktop:

1. Install Claude Desktop and the @anthropic-ai/dxt npm package.
2. Clone the Windows-MCP repository from GitHub.
3. Navigate into the directory and run npx @anthropic-ai/dxt pack to build the extension.
4. In Claude Desktop, go to Settings > Extensions > Advance Settings > Install Extension and select the .dxt file you just created.

For Perplexity Desktop:

1. Install Perplexity Desktop and clone the repository.
2. In the app, head to Settings > Connectors > Add Connector > Advanced.
3. Name it Windows-MCP and paste the provided JSON configuration, making sure to update the path to the Windows-MCP directory.

For Gemini CLI:

1. Install the Gemini CLI and clone the repo.
2. Open the settings.json file located in your %USERPROFILE%/.gemini folder.
3. Add the windows-mcp server configuration, again ensuring the directory path is correct.

For Qwen Code:

1. Install Qwen Code and clone the repository.
2. Find the settings.json file in %USERPROFILE%/.qwen.
3. Add the windows-mcp server configuration with the correct path.

For Codex CLI:

1. Install the Codex CLI and clone the project.
2. Navigate to %USERPROFILE%/.codex/config.toml.
3. Add the windows-mcp configuration block with the proper directory path.

3. Once installed, the MCP server exposes a variety of tools to the AI client:

• Click-Tool: Clicks the mouse at specified screen coordinates.
• Type-Tool: Enters text, with an option to clear the field first.
• Clipboard-Tool: Performs copy and paste actions using the system clipboard.
• Scroll-Tool: Scrolls a window or a specific area vertically or horizontally.
• Drag-Tool: Drags the mouse from a starting point to an ending point.
• Move-Tool: Moves the mouse pointer to a specific location on the screen.
• Shortcut-Tool: Executes keyboard shortcuts like Ctrl+C or Alt+Tab.
• Key-Tool: Presses a single keyboard key.
• Wait-Tool: Pauses execution for a specified amount of time.
• State-Tool: Captures a comprehensive snapshot of the desktop, including active apps, UI elements, and a screenshot.
• Resize-Tool: Changes the size or position of an application window.
• Launch-Tool: Opens an application from the start menu.
• Shell-Tool: Executes PowerShell commands.
• Scrape-Tool: Extracts all textual information from a webpage.

FAQs

Q: Does this work on non-English versions of Windows?
A: It’s highly recommended to use English as the default language. Some tools like Launch-Tool and Switch-Tool might not work as expected with other languages unless you disable them in the server configuration.

Q: What are the main limitations right now?
A: Currently, it has trouble selecting specific text within a paragraph because it relies on the accessibility tree. Also, the Type-Tool isn’t ideal for writing code in an IDE as it pastes the entire block at once. And no, you can’t use it to play video games.

Q: Is it secure to let an AI control my desktop?
A: You should exercise caution. The tool gives an AI agent direct control over your system to perform actions. It’s best to use it in controlled environments and avoid tasks where mistakes could cause problems.

Q: Can I create my own custom tools?
A: Yes. You can look at the existing tools as a reference and build your own to suit your specific automation needs.