Computer Use MCP
Enables LLMs to control Windows desktops via local OCR, optional YOLO icon detection, and mouse/keyboard actions.
README
Computer Use MCP
Windows Computer Use MCP: local OCR (PP-OCRv4 / RapidOCR) + optional YOLO icon detect + mouse/keyboard.
You do not need the full Enikk desktop app — only this MCP layer.
License: PolyForm Noncommercial 1.0.0 — free for non-commercial use; commercial use requires a paid license (COMMERCIAL.md).
Third-party / Enikk MIT notes: NOTICE.
Platform: Windows 10/11 only. Requires uv (
uvx).
Install with uvx (from GitHub)
uvx --from git+https://github.com/Just1dream/computer-use-mcp computer-use-mcp-auto
Cursor / Claude mcp.json
{
"mcpServers": {
"computer-use": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/Just1dream/computer-use-mcp",
"computer-use-mcp-auto"
]
}
}
}
First run downloads Python dependencies into an isolated uv environment (needs network). Later runs use the cache.
Optional env
| Variable | Meaning |
|---|---|
COMPUTER_USE_WEIGHTS |
Folder with rapidocr/ and icon_detect/model.onnx |
COMPUTER_USE_HOME |
Portable install root (set by run-mcp.cmd) |
COMPUTER_USE_SKIP_BOOTSTRAP=1 |
Skip copy/download of weights on start |
COMPUTER_USE_SCREENSHOT_DIR |
Where PNGs are saved |
COMPUTER_USE_MAX_DIM |
Max screenshot side (default 1366) |
Weights are not in this repo (large). Place them under %LOCALAPPDATA%\ComputerUseMCP-Auto\weights or set COMPUTER_USE_WEIGHTS. OCR may fall back to RapidOCR defaults if models are missing; icon YOLO needs icon_detect/model.onnx.
Portable zip (offline-friendly)
On a Windows machine with the repo:
.\pack-zip.ps1 # slim
.\pack-zip.ps1 -IncludeWeights # embed models if available locally
Recipients: unzip → .\install-to-cursor.ps1 → restart Cursor. See FOR_RECIPIENTS.md.
Tools
| Tool | Description |
|---|---|
list_windows / find_window / focus_window |
Window discovery |
screenshot / parse_ui |
Capture + local OCR/icons ([0,1000] coords) |
click_normalized / type_text / hotkey / press_key / scroll_normalized |
Input |
weights_info / bootstrap_weights |
Model path / ensure weights |
Typical loop: list_windows → parse_ui / screenshot → click_normalized.
Security
This MCP can see the screen and control mouse/keyboard. Only enable for trusted users/hosts.
License
- Default: PolyForm Noncommercial 1.0.0 — non-commercial free; commercial use needs a separate license (
COMMERCIAL.md). - Upstream: Enikk MIT portions and third-party packages — see
NOTICE.
Commercial inquiries: edit the email in COMMERCIAL.md (placeholder YOUR_EMAIL@example.com).
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。