WinSight MCP
Windows Screen Capture MCP Server — give Claude Code eyes on your Windows desktop.
README
WinSight MCP
mcp-name: io.github.TheoEwzZer/winsight
Windows Screen Capture MCP Server — give Claude Code eyes on your Windows desktop.
WinSight is an MCP server that lets Claude Code capture your screen, manage windows, and launch applications on Windows.
Features
- Screenshot the full screen, a specific region, or a specific window
- Window capture uses Win32
PrintWindowAPI — captures the real window content even when it's behind other windows - List and inspect open windows and monitors (title, position, size, state, resolution)
- Control windows — move, resize, minimize, maximize, restore, and focus
- Launch applications and wait for their windows
Requirements
- Windows 10/11
- Python 3.10+
Quick Start
Option 1: uvx (recommended)
No install needed — runs directly:
{
"mcpServers": {
"winsight": {
"command": "uvx",
"args": ["winsight-mcp"]
}
}
}
Add this to your project's .mcp.json or ~/.claude/claude_desktop_config.json.
Option 2: pip install
pip install winsight-mcp
Then configure:
{
"mcpServers": {
"winsight": {
"command": "winsight-mcp"
}
}
}
Option 3: From source
git clone https://github.com/TheoEwzZer/WinSight-MCP.git
cd WinSight-MCP
uv sync
{
"mcpServers": {
"winsight": {
"command": "uv",
"args": ["--directory", "/path/to/WinSight-MCP", "run", "winsight-mcp"]
}
}
}
Tools
Screenshot
| Tool | Description |
|---|---|
take_screenshot |
Capture the full screen or a specific monitor |
screenshot_window |
Capture a specific window by title (works even if behind other windows) |
screenshot_region |
Capture a rectangular region of the screen |
Window Management
| Tool | Description |
|---|---|
list_windows |
List all visible windows with optional title filter |
get_window_info |
Get detailed info about a window (position, size, state) |
focus_window |
Bring a window to the foreground |
resize_window |
Resize a window to specific dimensions |
move_window |
Move a window to a specific position |
minimize_window |
Minimize a window to the taskbar |
maximize_window |
Maximize a window to fill the screen |
restore_window |
Restore a minimized or maximized window to its normal state |
wait_for_window |
Wait for a window to appear (adaptive polling with timeout) |
System
| Tool | Description |
|---|---|
list_monitors |
List all monitors with resolution, position, and primary flag |
open_application |
Launch an application and optionally wait for its window |
Examples
Once the MCP server is connected, you can ask Claude Code things like:
- "Take a screenshot of my screen"
- "List all open windows"
- "Capture the Notepad window"
- "Open calculator and take a screenshot of it"
- "Focus the Chrome window"
- "Resize the app window to 800x600 and take a screenshot"
- "Move the window to the top-left corner"
- "What monitors do I have?"
Testing
The project has 112 tests covering all modules. Tests use mocks for Win32 APIs so they run on any platform.
Running tests
uv run pytest
Test structure
tests/
conftest.py # Shared fixtures and Win32 stubs
test_types.py # TypedDict definitions validation
test_screenshot.py # Screen/region/window capture (mss, Win32 DC)
test_window_manager.py # Window listing, find, focus, resize, move, min/max/restore
test_process_manager.py # Application launch and window polling
test_server.py # MCP tool registration and integration
Adding tests
- Put new tests in the matching
test_<module>.pyfile - Use the shared fixtures from
conftest.py(sample_window_info,mcp_server,fake_png_bytes) - Mock Win32 APIs with
@patch("winsight_mcp.<module>.win32gui")— never call real Win32 functions in tests - For server integration tests, use the
_callhelper to invoke tools and_textto extract string results
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。