# Windows-MCP

> Let AI agents operate Windows: open apps, control windows, click and type, and read UI state.

## Facts
- **Category:** [MCP Servers](https://agentsearchengine.app/category/mcp-servers)
- **Type:** MCP server
- **Pricing:** Open source
- **GitHub stars:** 7,114
- **Language:** Python
- **License:** MIT
- **Last repo activity:** 1 day ago (Sep 21, 2026)
- **Official site:** https://github.com/CursorTouch/Windows-MCP
- **Source code:** https://github.com/CursorTouch/Windows-MCP
- **Listing on Agent Search Engine:** https://agentsearchengine.app/agents/windows-mcp
- **Data verified:** September 2026

## Best for
Automating desktop applications and UI testing on Windows from an MCP client.

## Avoid if
You're on macOS or Linux; it supports Windows 7 through 11 only.

## About Windows-MCP
Windows-MCP is a lightweight MCP server for computer use on Windows. It interacts with native UI elements, launches and controls applications, simulates keyboard and mouse input and captures window and UI state, and works with any LLM because it does not rely on computer-vision models. It runs over stdio by default, with SSE and streamable HTTP transports plus authentication and IP allowlists for remote use.

## Connect to an MCP client
Launch command from the project's README: `uvx windows-mcp serve`

### Claude Code (Run in your terminal)

```shell
claude mcp add windows-mcp -- uvx windows-mcp serve
```

### Cursor (~/.cursor/mcp.json or .cursor/mcp.json)

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

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "windows-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": ["windows-mcp", "serve"]
    }
  }
}
```

### Claude Desktop (claude_desktop_config.json)

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

### Codex (~/.codex/config.toml)

```toml
[mcp_servers.windows-mcp]
command = "uvx"
args = ["windows-mcp", "serve"]
```

## Alternatives
Windows-MCP sits in mcp servers. Related tools indexed here:
- [MCP Reference Servers](https://agentsearchengine.app/agents/mcp-reference-servers) — The MCP steering group's reference servers: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time. · Open source · MCP server · 91k★ · TypeScript · MIT AND Apache-2.0 · active today
- [context7](https://agentsearchengine.app/agents/context7) — Context7 Platform -- Up-to-date code documentation for LLMs and AI code editors · Open source · MCP server · 62k★ · TypeScript · MIT · active today
- [chrome-devtools-mcp](https://agentsearchengine.app/agents/chrome-devtools-mcp) — Chrome DevTools for coding agents · Open source · MCP server · 53k★ · TypeScript · Apache-2.0 · active today
- [Playwright MCP](https://agentsearchengine.app/agents/playwright-mcp) — Expose Playwright browser tools to MCP clients using accessibility snapshots. · Open source · MCP server · 37k★ · TypeScript · Apache-2.0 · active 4 days ago
- [github-mcp-server](https://agentsearchengine.app/agents/github-mcp-server) — GitHub's official MCP Server · Open source · MCP server · 33k★ · Go · MIT · active today
- [serena](https://agentsearchengine.app/agents/serena) — A powerful MCP toolkit for coding, providing semantic retrieval and editing capabilities - the IDE for your agent · Open source · SDK / library · 30k★ · Python · MIT · active today
- [MCP for Blender (blender-mcp)](https://agentsearchengine.app/agents/blender-mcp) — Connect Blender to Claude or any LLM client for prompt-driven 3D modelling and scene editing. · Open source · MCP server · 29k★ · Python · MIT · active 2 days ago
- [fastmcp](https://agentsearchengine.app/agents/fastmcp) — The fast, Pythonic way to build MCP servers and clients. · Open source · SDK / library · 28k★ · Python · Apache-2.0 · active today

---
Source: [Agent Search Engine](https://agentsearchengine.app) — an independent, hand-curated index of AI agents, MCP servers, and frameworks. Rankings are never sold; sponsored placements are labeled and never numbered. Full directory: https://agentsearchengine.app/llms-full.txt