mcp-windows by sbroenne - MCP Server

🪟 Windows MCP Server

Windows automation that actually works. Uses the Windows UI Automation API to find buttons by name, not pixels. Tested with real AI models before every release.

Why This Exists

Screenshot-based automation doesn't work reliably. Vision models guess wrong, coordinates break when windows move or DPI changes, and you burn through thousands of tokens on retry loops. We tried it (check the commit history) — it failed too often to be useful.

Windows MCP Server asks Windows directly: "What buttons exist in this window?" Windows knows. It's deterministic.

How It Works

# 1. Find the window
window_management(action='find', title='Notepad') → handle='123456'

# 2. Click elements by name
ui_click(windowHandle='123456', nameContains='Save')

# 3. Type into fields
ui_type(windowHandle='123456', controlType='Edit', text='Hello World')

# 4. Fallback for games/canvas — screenshot + mouse
screenshot_control(windowHandle='123456') → element coordinates
mouse_control(action='click', x=450, y=300)

Same command works every time. Any machine. Any DPI. Any theme.

Key Features

🧠 Semantic UI — Find elements by name, not coordinates. Works regardless of DPI, theme, or window position.
� Multi-Monitor — Full support for multiple displays with per-monitor DPI scaling.
🧪 LLM-Tested — 54 tests with real AI models (GPT-4.1, GPT-5.2). 100% pass rate required for release.
💻 Broad App Support — Tested against classic Windows apps, modern Windows 11 apps, and Electron apps (VS Code, Teams, Slack).
🔄 Full Fallback — Screenshot + mouse + keyboard for games and custom controls.
🪙 Token Optimized — Short property names, JPEG screenshots, auto-scaling. ~60% fewer tokens than standard JSON.

Installation

VS Code Extension — Install from Marketplace. Works with GitHub Copilot automatically.

Standalone — Download from Releases. Add to your MCP config:

{ "servers": { "windows": { "command": "path/to/Sbroenne.WindowsMcp.exe" } } }

Tools

Tool	Purpose
`ui_click`	Click buttons, checkboxes, menu items by name
`ui_type`	Type into text fields
`ui_find`	Discover elements in a window (with timeout/retry)
`ui_read`	Read text from elements (with OCR fallback)
`ui_file`	Handle Save dialogs
`screenshot_control`	Get element metadata (image optional)
`window_management`	Find, activate, move, resize windows
`mouse_control`	Coordinate-based clicks (fallback for games)
`keyboard_control`	Hotkeys and key sequences
`app`	Launch applications

Full reference:

⚠️ Caution

This MCP server controls your Windows desktop. Use responsibly.

Testing

dotnet test                                      # All tests
dotnet test --filter "FullyQualifiedName~Unit"   # Unit only

Framework coverage: Tests run against WinForms, WinUI 3, and Electron apps.

LLM tests: 54 tests with real AI models (GPT-4.1, GPT-5.2). 100% pass rate required for release.

cd tests/Sbroenne.WindowsMcp.LLM.Tests
.\Run-LLMTests.ps1

Requires Azure OpenAI access. See .

Related Projects

agent-benchmark — LLM agent testing framework (powers our integration tests)
Excel MCP Server — AI-powered Excel automation
OBS Studio MCP Server — AI-powered streaming control

Documentation

Document	Description
	Complete tool reference — all actions, parameters, examples
	Build instructions, coding guidelines, PR process
	How to run LLM integration tests
	Azure OIDC and GitHub Actions configuration

License

MIT — see

Contributing

See