Skip to content
nuphus-mcp

nuphus-mcp

by Nuphus Team
Released
15 views
MIT
Repository
Developer Tools

Summary

Computer use for any MCP client: 36 tools covering screenshots, window control, mouse and keyboard, and Chrome over CDP, with local OCR built in and vision on your own key.

Features

  • 36 MCP tools — 15 desktop, 21 browser
  • Screenshots, window activate/move/resize, mouse, keyboard, clipboard
  • Chrome automation over the DevTools Protocol
  • Local OCR with PaddleOCR plus optional YOLO icon detection — no key needed
  • Vision through your own OpenAI-compatible or Anthropic-native model (BYOK)
  • Single Rust binary over stdio — no daemon, no network service
  • --confirm-write gating and a documented threat model with per-tool safety annotations

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it.
  3. Restart the agent — the "nuphus-mcp" tools will be available.
{
  "mcpServers": {
    "nuphus-mcp": {
      "command": "nuphus-mcp",
      "args": [
        "--confirm-write"
      ]
    }
  }
}

Description

nuphus-mcp is a cross-platform desktop-automation server that gives any MCP client the ability to see and drive the machine it runs on. It speaks JSON-RPC 2.0 over stdio — one binary, no daemon, no network service — so Claude Desktop, Cursor, VS Code, Copilot or any other client can connect and immediately control the screen, windows, keyboard, mouse and Chrome.

The 36 tools

15 desktop tools: screen size, screenshot as PNG or base64, window listing, per-window activate / screenshot / move / resize / info, mouse click, drag, scroll and position, keyboard input and hotkeys, clipboard write and clear. These are implemented on the desktop-api crate over xcap and Win32, with no Tauri dependency.

21 browser tools driving Chrome through the DevTools Protocol via the nuphus-browser crate.

Seeing the screen

Two complementary tools handle perception. desktop_perceive runs local OCR with PaddleOCR — models download on first run, with optional YOLO icon detection — so text on screen is readable without sending anything anywhere. desktop_vision sends a screenshot to your own vision model over an OpenAI-compatible or Anthropic-native API. Used together they give an agent both semantic understanding and precise coordinates.

Desktop and browser automation need no API key at all; only the vision tool does, and it is bring-your-own.

Safety

Write operations can be gated behind --confirm-write, or the equivalent environment variable so one setting covers every MCP client on the machine. The repository documents a threat model and per-tool safety annotations — worth reading before you hand an agent your mouse.

Written in Rust, MIT licensed, and mirrored on Gitee with Chinese documentation for access from mainland China.

Related MCP Servers

MCP: Headroom

by Headroom Labs

New

Context-compression MCP server that shrinks tool outputs, logs, files and RAG chunks before they reach the model, cutting 60-95% of tokens on JSON payloads.

Developer Tools
2 views

MCP: Engram

by Gentleman Programming

New

Persistent memory for coding agents as a single Go binary — SQLite and FTS5 behind MCP, an HTTP API, a CLI and a TUI, with no Node, Python or Docker required.

Developer Tools
2 views

MCP: agent-device

by Callstack

New

Give a coding agent a live feedback loop on real mobile, TV and desktop apps — inspect, tap, type and capture evidence on iOS, Android and HarmonyOS over MCP or a CLI.

Developer Tools
3 views

MCP: Weights & Biases MCP Server

by Weights & Biases

New

The official W&B MCP server: query experiment runs, Weave LLM traces, artifacts and registries in natural language, and write findings back as a W&B report.

Developer Tools
3 views
Browse all MCP servers →