Windows-MCP
Windows-MCP is a lightweight, open-source MCP server that connects AI agents to the Windows operating system. It interacts with native Windows UI elements, opens apps, controls windows and simulates user input. It works with any LLM and does not need computer-vision techniques or a specially fine-tuned model, and the README gives typical latency between actions of 0.7 to 2.5 seconds depending on load and model speed. Supported systems are Windows 7, 8, 8.1, 10 and 11, and it installs from PyPI so it runs with `uvx windows-mcp`.
The tools are Click, Type, Scroll, Drag, Move, Shortcut, Wait, State (a combined snapshot of active apps and interactive, textual and scrollable elements with a desktop screenshot), App (launch, resize, move and switch windows), Shell (execute PowerShell commands) and Scrape (read a whole web page), plus a DOM mode for browser automation. The README notes it has been distributed as a Claude Desktop extension and listed in the MCP Registry, and it states plainly that it operates with full system access and can perform irreversible operations. Anonymous usage telemetry (no tool arguments or outputs) is on by default and can be disabled with ANONYMIZED_TELEMETRY=false. It is MIT licensed.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-25
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 4
- Latest release
- v0.8.8 (2026-10-10)
- Open issues
- 30
- Recent issues with at least one comment
- 5/17
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #236 Add optional service mode to see and handle UAC prompts (Streamline and empower autonomous agent workflows on Windows)11 comments · 2026-05-12
- #29 Can't recognize element in sheet.7 comments · 2025-08-17
- #332 WatchDog UIA focus listener crashes the whole MCP server after long uptime (COM "Catastrophic failure", no Python traceback)3 comments · 2026-07-13
- #412 Startup crash when Claude Desktop launches several instances at once: comtypes.gen race on UIAutomationCore wrapper (0.7.2)2 comments · 2026-09-12
- #401 Claude Desktop extension channel still ships v0.7.2 without the #358 comtypes cache-race fix - startup race still killing servers2 comments · 2026-08-31
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You need an AI agent to perform real Computer Use on Windows (mouse/keyboard simulation, window management, and application interaction).
- You want to run automated UI QA testing or desktop task macros on Windows 10/11 with zero complicated setup via `uvx windows-mcp`.
- Your tasks are strictly confined within web browsers; Playwright-MCP or Browserbase are faster, headless, and far more robust for DOM tasks.
- You are running on macOS (where OpenClaw companion apps fit better) or headless Linux servers (where CUA or OSWorld sandboxes are designed).
How it differs
vs playwright-mcp★ 38.7kWindows-MCP automates the entire OS desktop (native Win32/WPF apps, File Explorer); Playwright-MCP is specialized for browser DOM automation with headless reliability.
vs chrome-devtools-mcp★ 54.3kWindows-MCP controls OS-level inputs without browser runtime introspection; Chrome-DevTools-MCP provides direct access to network traffic, DOM, and console logs.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!