Windows-MCP is an open-source MCP server that gives AI agents computer-use abilities on Windows, letting language models see the screen and operate the desktop like a person would. Published on GitHub under the MIT license, it has earned roughly 9,800 stars and is under very active development, with updates continuing through October 2026.
Windows-MCP implements the Model Context Protocol for the Windows desktop. Once connected to an MCP-compatible client such as a coding assistant or agent framework, the model can capture screenshots, inspect the UI tree, move the mouse, click, type, and interact with arbitrary Windows applications. That closes the loop for computer-use agents on the platform most office software runs on: instead of being limited to APIs and terminals, an agent can drive legacy desktop apps, browsers, and system dialogs directly. The server is written in Python and distributed via PyPI, and it targets Windows 10 and 11. Because it acts through the real desktop, it works with software that has no automation API at all, which makes it popular for RPA-style experiments and agent benchmarks.
Windows-MCP is free and open source under the MIT license. You install and run the server locally at no cost; the only expenses are whatever model API or local inference costs your chosen AI client incurs while driving the desktop.
Windows-MCP is for developers building computer-use agents, automation engineers prototyping RPA flows on Windows, and researchers evaluating agent performance on real desktop tasks. If your workflow involves Windows-only software and you want an LLM to operate it end to end, this server is the bridge.