convez.ai
MCP ServerAutomaçãoAvançado
sandraschi

Devices MCP

sandraschi/devices-mcp

MCP server + web dashboard para controlar dispositivos IoT da casa

Ver no GitHub
BomPluginScore · de 0 a 100

Stack de IoT para casa conectada que integra dispositivos LAN (Tapo, Hue, Ring, Nest, Shelly) como ferramentas MCP para agentes Claude, web dashboard React e aplicativo Windows desktop. Fornece controle centralizado de câmeras, energia, iluminação e sensores via dashboard web ou como tools nativos para automação com IA.

Instalação

$npx convez-ai sandraschi/devices-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
0
Acompanhando
0
Saúde
100%
Último commit
há 2 semanas
Versão
v2.4.1
Issues abertas
4
Contribuidores
2

Casos de uso

  • Integrar câmeras Tapo e USB com agentes Claude para análise visual
  • Monitorar plugs de energia e controlar iluminação Philips Hue via MCP
  • Dashboard unificado para status de dispositivos LAN (Ring, Nest, Shelly)
  • Automação de casa inteligente com FastMCP para aplicações conversacionais

Veredito

Projeto beta focado em automação de IoT doméstica via MCP, com bom suporte para múltiplos fabricantes. Excelente para quem quer integrar Claude em stack de casa inteligente, mas ainda em fase experimento (expect rough edges, cold starts lentos e mudanças de API).

README

A tradução para o português deste README ainda está na fila.

devices-mcp

Beta home IoT stack for Sandra's fleet: MCP tools for agents, a web dashboard for humans, and an optional Windows desktop installer that bundles both. LAN devices (Tapo cameras/plugs, Philips Hue, Ring, Nest via Home Assistant, Shelly, USB cams, Netatmo, Vienna presets) are configured in config.yaml — not shipped as secrets.

Not production-hardened. Expect rough edges, slow cold starts (~2 min for full PyInstaller sidecars), and API changes between releases.


Table of contents

SectionLink
Three delivery legsThree legs
Features (realistic)Features
Quick installQuick install
DocumentationDocumentation
PortsPorts
RequirementsRequirements
LicenseLicense

Three delivery legs

LegWhat you getBest for
1 — MCP serverFastMCP stdio/HTTP, 40+ portmanteau toolsClaude Desktop, Cursor, automation
2 — WebappVite/React UI + FastAPI (web-sota/)Browser dashboard, same APIs as desktop
3 — Desktop (Tauri)Devices-MCP-*-setup.exe (~247 MB)Double-click Windows install; spawns camera + backend sidecars

Use one backend on port 10717 at a time. If you already run the stack via NSSM or just serve, open the dashboard in a browser — do not start a second backend from the desktop app.

Details: docs/ARCHITECTURE.md · Desktop: docs/DESKTOP.md


Features

  • Cameras: Tapo/ONVIF, USB webcam/microscope (OpenCV), streaming hooks (quality varies by device/driver).
  • Energy: Tapo P115 plugs; LAN discovery optional when cloud creds are set.
  • Lighting: Philips Hue (incl. HomeAware on supported bridges), Tapo lighting.
  • Security / sensors: Ring, Nest Protect via Home Assistant, Shelly, Netatmo, Open-Meteo (Vienna default).
  • Dashboard: Status table with device health, per-domain pages (cameras, energy, weather, robots, …).
  • MCP: Portmanteau tools (camera, energy, lighting, ring, nest, weather, …) on FastMCP 3.2.
  • Fleet priority: GET /api/fleet/priority — home-safety incidents for Fritz urgent dispatch (kitchen temp, CO, smoke, Ring).

Not included: guaranteed uptime SLA, cloud hosting, plug-and-play without config.yaml, or a single 12 MB exe (full desktop bundle is ~247 MB with embedded Python sidecars).


Quick install

Windows desktop (leg 3): GitHub Releases → Devices MCP_*_x64-setup.exe (single NSIS installer — that's all you need).

MCP bundle (leg 1): Releases → devices-mcp.mcpb → Claude Desktop.

Developers (leg 1+2): see INSTALL.md.

git clone https://github.com/sandraschi/devices-mcp
cd devices-mcp
uv sync
.\web-sota\start.ps1
# http://127.0.0.1:10717/app/

Build & Test (NSIS installer)

The desktop installer bundles TWO sidecars (backend + camera helper) into a single Devices MCP_*_x64-setup.exe (~247 MB).

just build-native      # Full pipeline: frontend → PyInstaller → Rust → NSIS
just cua-nsis-test     # Install → CUA test → screenshot → uninstall

The CUA smoke test (scripts/cua-smoke.py) verifies:

  1. Silent install succeeds
  2. Backend health responds 200 on port 10717
  3. Window renders at correct size
  4. Diagnostics endpoint reports all systems
  5. Clean uninstall

See CUA-NSIS Smoke Testing — fleet standard.

Prerequisites: Rust toolchain, MSVC Build Tools, Tauri CLI 2.x.


Documentation

DocContents
INSTALL.mdAll install paths (Tauri, MCPB, clone, NSSM)
docs/README.mdFull documentation index
docs/ARCHITECTURE.mdMCP + webapp + desktop, ports, sidecars
docs/DESKTOP.mdTauri installer, splash, NSSM coexistence
docs/CONFIGURATION.mdconfig.yaml, Vienna preset, discovery flags
docs/TOOLS.mdMCP tool families (summary)
docs/DEVELOPMENT.mdjust, build scripts, contributing
docs/TROUBLESHOOTING.mdWhite screen, logs path, firewall
docs/FLEET_INTEGRATION.mdFritz /api/fleet/priority consumer contract
CHANGELOG.mdVersion history

LLM index: llms.txt · full: llms-full.txt


Ports

PortService
10717FastAPI backend (API + /app/ SPA)
10716Vite dev frontend only
10715USB camera helper (optional)

Requirements

  • OS: Windows 10/11 x64 for desktop installer; dev backend also runs on Linux/macOS with reduced hardware support.
  • Config: config.yaml with your LAN IPs and credentials (see config.example.yaml, home_preset: vienna template).
  • Network: Devices on reachable LAN; mDNS/LAN broadcast for Tapo discovery optional.

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados