Scry — IA para Robôs ROS 2
phaneron-robotics/scry
Depure e converse com robôs ROS 2 pelo celular
Ver no GitHubMCP server que expõe o grafo ROS 2 como 119 ferramentas IA-callable, permitindo depuração e monitoramento de robôs via assistente IA em tempo real, sem nuvem. Roda localmente na rede privada do usuário, com a app Android como cliente espesso que roteia requisições entre IA e robot.
Instalação
npx convez-ai phaneron-robotics/scry1 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
Casos de uso
- Depuração de tópicos, nós e serviços ROS 2 em campo
- Monitoramento de robôs durante demonstrações e testes
- Diagnóstico rápido de problemas sem laptop dedicado
Veredito

Solução robusta e bem-arquitetada que democratiza o debugging de robôs ROS 2 via IA local. Integração MCP + rclpy + Android é técnica, mas o fluxo final é acessível mesmo para não-especialistas em robótica.
README
A tradução para o português deste README ainda está na fila.
Scry
Your ROS 2 robot, in your pocket. Debug, monitor, and talk to any ROS 2 system from your Android phone, through an on-device AI assistant.
Docs · Quick start · How it works · Demos
What is Scry?
Debugging a ROS 2 robot usually means crouching at a laptop with rviz, rqt, and a wall of ros2 topic echo terminals. That's fine at your desk, but painful when you're walking beside the robot, demoing it, or out in the field. Scry puts that whole workflow in your pocket: you ask a question in plain English, like "why isn't /cmd_vel publishing?", and an on-device AI assistant inspects your robot's topics, nodes, services, parameters, and diagnostics live over your own network, then answers with structured panels, plots, and one-tap actions. No cloud backend, no telemetry. Your AI key, your robot, your network.
How it works
flowchart LR
User(["You\ntext · voice · image"])
subgraph Phone["Scry app, Android"]
AI["AI assistant\n+ MCP routing\n+ rich rendering"]
end
subgraph Robot["Your robot"]
Connect["scry-connect\nMCP server"]
ROS["ROS 2 graph\nany DDS / RMW"]
end
Cloud(["AI provider\nOpenRouter · Ollama"])
User <--> AI
AI <-->|"your Wi-Fi"| Connect
Connect <-->|"rclpy"| ROS
AI <-.->|"BYO key · or fully local"| Cloud
classDef brand fill:#292826,stroke:#A3B86C,stroke-width:1px,color:#E8E4D9
class Phone,Robot,AI,Connect,ROS brand
The phone is the thick client. It runs the AI loop, routes tool calls, renders the results, keeps your background monitors running, and manages your fleet. The robot just runs scry-connect, a small Python MCP server that exposes its ROS 2 graph as 119 AI-callable tools. Reads are free; anything that changes the robot asks for your approval first. Because cloud AI APIs can't reach your private Wi-Fi, the phone proxies every tool call between the AI and the robot, so nothing routes through a Scry server.
Quick start
About fifteen minutes, end to end. Full walkthrough in the Get started guide.
1. Get the app
Get Scry on Google Play →
2. Install scry-connect on the robot
pip (simplest):
pip install scry-connect
source /opt/ros/$ROS_DISTRO/setup.bash
scry-connect
Docker (no Python setup, multi-arch, works on Jetson / Pi):
docker run --rm --network host \
ghcr.io/phaneron-robotics/scry-connect:jazzy
It binds 0.0.0.0:5339, LAN-only, and prints a pairing QR on startup.
3. Pair & ask
Scan the QR in the app, add an AI provider, and ask your first question:
"What's my robot's health?"
Install on Android · Install scry-connect · Pair the phone & robot · First session
AI providers
Bring your own key, or run fully offline. Provider and model are picked from the chat top-bar chip.
| Provider | Notes |
|---|---|
| OpenRouter | Recommended. One key unlocks 300+ models (Claude, GPT, Gemini, Llama, and more) and has a free tier. |
| Ollama | Run a model locally, on your own machine. No cloud, no key, fully offline. |
Scry sends inference through whichever you choose; it stores only your credentials, never your robot or chat data.
Demos
Demo GIFs and videos coming here: chat-driven debugging, voice queries, live sensor panels, background monitors, and fleet view.
What you can do
- Debug by chatting. Ask "why isn't
/cmd_velpublishing?" and Scry inspects topics, nodes, services, and parameters live and explains what it found. - Voice & images. Talk hands-free, or attach a screenshot and ask "what's wrong in this scene?" Transcription runs on your phone.
- Act with one tap. Publish a topic, set a parameter, call a service, or drive a lifecycle change. Every write shows exactly what it'll do and waits for your approval.
- Background monitors. Say "alert me if
/odomdrops below 10 Hz" and Scry watches, then pings you the moment a condition trips. - Live panels & plots. Sensor readouts, scene snapshots, transform trees, and live plots render right in the chat.
- Viz tab. A dedicated long-running surface for a 3D scene view, camera feeds, behavior trees, a geomap, live plots, sensor panels, bag playback, and teleop.
- Fleet view. Monitor and compare multiple robots from one screen.
Documentation
Everything lives at phaneron-robotics.github.io/scry:
- Get started: Overview · Android · scry-connect · Choose your AI · Pair · First session
- Use Scry: Chat · Attachments · Voice · Monitors · Remote
- How it works: Architecture
- Reference: MCP tools · Permissions
- Legal: Privacy · Data safety · Security
Acknowledgements
Scry is built on the shoulders of great open source:
- ROS 2 & rclpy, the robot graph Scry speaks to.
- Model Context Protocol, the tool-calling standard
scry-connectimplements. - OpenRouter & Ollama, the inference layer Scry routes through (cloud or fully local).
- Jetpack Compose, Hilt, OkHttp, Room, the Android stack.
License
Apache-2.0.
Made with 💚 by the Phaneron Robotics team.