Connect a robot

Robots connect through an agent on the computer attached to them. The agent finds the robot's cameras and sensors, registers them with Scope over MCP and starts a small bridge that streams them here.

Connect an agent on the robot’s computer
Open your agent harness on the machine the robot is plugged into and add this Scope’s MCP server.
MCP server · ~/.gemini/antigravity-cli/mcp.json
{ "mcpServers": { "sceneagent": { "serverUrl": "https://sceneagent.dev/mcp", "headers": { "Authorization": "Bearer <your sa_ token>" } } } }
Full agent setup →
Ask it to connect the robot
Paste this into the agent. It decides which feeds the robot has; you can rename or change them afterwards.
Robot name optionalShare with
Prompt
On this computer, which is attached to my robot, connect to the SceneAgent MCP server at /mcp. Inspect the robot's hardware: every camera (USB, RTSP or other) and every sensor stream such as arm joint states, grippers, force/torque and IMUs. Ask me about anything you cannot identify. Call register_robot, the embodiment, and one feed per camera and sensor, kept private to me. Run scripts/scope-robot-bridge.py with the ingest key it returns, mapping each camera feed to its device and piping sensor readings into the bridge as JSON lines. Call get_robot until the robot is online and every feed is live, then open its browserUrl.
Watch it come online
Robots the agent registers while this page is open appear here, then go live as the bridge starts streaming.
Waiting for the agent to register a robot · checking every 2 s