An MCP server that lets AI agents interact with the Windows operating system via the accessibility tree, enabling UI automation, file management, and PowerShell execution without relying on computer vision. It runs locally as a server component, while LLM clients may operate remotely.
Project overview
It provides an accessibility-tree-based approach to desktop automation, allowing any LLM to control Windows applications, manage files, and execute commands without fine-tuned vision models.
Project type
MCP
Deployment
Refer to project documentation
License
MIT
Best for
Automating Windows desktop tasks and controlling native applications via MCP-compatible clients.
Managing files, running PowerShell, and performing browser automation through an AI agent.
Key capabilities
Interacts natively with Windows UI elements, opens apps, controls windows, and simulates user input.
Includes tools for basic keyboard, mouse operation and capturing window/UI state.
Provides a special use_dom=True mode for State-Tool that focuses exclusively on web page content, filtering out browser UI elements. Supports Chrome, Edge, and Firefox.
Provides tools for MCP clients to click, type, scroll, take screenshots, manage files, run PowerShell, scrape webpages, and read/write registry.
Supports authentication, IP allowlists, TLS/HTTPS, OAuth 2.0 + PKCE, and CORS origin management for secure network access.
Limitations and risks
Selecting specific sections of text in a paragraph is not supported due to reliance on the accessibility tree.
Type-Tool is not meant for programming in an IDE because it types the program as a whole.
Cannot be used to play video games.
Getting started
Setup is rated as easy. Install prerequisites like uv, utilize simple config file setup, and run the server directly with uvx windows-mcp serve.
Alternatives and comparisons
A fair-code platform to build and deploy AI agents and workflows with a visual canvas, custom code, and 1500+ integrations.
An extensible, open-source AI workflow automation platform with a no-code builder where contributed pieces become MCP servers.
A lightweight, extensible AI assistant and agent harness that runs 24/7, supports multiple LLMs, and manages long-term memory.