Windows-MCP

MIT

windowscomputer-usedesktop-automationgui

Summary

Computer-use MCP server for Windows — 19 tools for clicking, typing, shell, filesystem and registry, driving the desktop without a vision model.

Features

19 tools: click, type, scroll, shortcut, PowerShell, filesystem, registry and more
Accessibility-tree driven — works without a vision model
DOM mode for Chrome, Edge and Firefox automation
Auth keys, IP allowlisting, TLS and OAuth 2.0
0.2–0.5s between actions

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

Open your agent's MCP configuration file.
Merge the snippet below into it.
Restart the agent — the "Windows-MCP" tools will be available.

{
  "mcpServers": {
    "windows-mcp": {
      "command": "uvx",
      "args": [
        "windows-mcp",
        "serve"
      ]
    }
  }
}

Claude Code

Add the server with a single CLI command.

claude mcp add windows-mcp -- uvx windows-mcp serve

Claude Desktop

Add the server to claude_desktop_config.json (Settings → Developer → Edit Config).

{
  "mcpServers": {
    "windows-mcp": {
      "command": "uvx",
      "args": [
        "windows-mcp",
        "serve"
      ]
    }
  }
}

Description

Most desktop automation for agents assumes a vision model looking at screenshots. Windows-MCP takes the accessibility route instead: it reads the UI tree, so a text-only model can click a button by name and know what is on screen without an image round-trip. Vision is supported but optional, which makes it usable with cheap models and keeps latency between actions at roughly 0.2–0.5 seconds.

Nineteen tools cover the surface: Screenshot and Snapshot for state, Click, Type, Scroll, Move, Shortcut for input, Wait and WaitFor for synchronisation, App for launching and switching, plus PowerShell, FileSystem, Scrape, MultiSelect, MultiEdit, Clipboard, Process, Notification and Registry. A DOM mode handles browser automation in Chrome, Edge and Firefox with page structure rather than pixels.

Because it is genuinely a remote-control surface, the security options matter and are present: auth keys, IP allowlisting, TLS and OAuth 2.0, configured by CLI flag or ~/.windows-mcp/config.toml.

The README is honest about limits, which is worth reading before wiring it into anything: selecting a specific range of text is unreliable, the Type tool is not a good fit for writing code in an IDE, and video games are out of scope. Requires Windows 7–11, Python 3.13+ and the uv package manager. MIT licensed.