mrpulor-gh/nuphus-mcp

View on GitHub ↗

Desktop automation MCP server — computer use for any AI agent: control screen, windows, mouse/keyboard, and Chrome via Model Context Protocol (stdio)

221 ★26 forksRustUpdated 18d ago

What you need to know

Lightweight cross-platform desktop automation MCP server - computer use for any AI agent with desktop and Chrome browser control over stdio

Install

npm install -g @nuphus/nuphus-mcp
nuphus-mcp

Usage

  • Control the screen, windows, mouse and keyboard plus drive Chrome via CDP
  • Pair desktop_vision (BYOK vision LLM) with desktop_perceive (local OCR) for pixel-precise computer use

Key features

  • 36 MCP tools (15 desktop + 21 browser)
  • Desktop automation: screenshots, window control, mouse/keyboard, clipboard
  • Local OCR via PaddleOCR with optional YOLO icon detection
  • Bring-your-own-key vision model over OpenAI-compatible or Anthropic native APIs
  • Zero-cost stdio - no daemon, no HTTP server, single binary
  • Optional strict-confirm mode for destructive tools

Best for

AI agents that need to physically operate a desktop and browser

Caveats

  • Windows recommended for full desktop control; macOS needs Accessibility permission; Linux desktop is partial
  • Browser tools require Chrome or Edge; vision tools require your own API key
  • Can physically control the machine - recommend running with --confirm-write
Platforms: Windows · macOS · LinuxClients: Claude Desktop · Cursor · VS Code · Copilot · Any MCP client

Reviewed 2026-08-11

Topics

Stars
221★
Forks
26
Language
Rust
License
MIT
Created
2026-08-01
Last push
2026-08-17