Wired Future
One action, two interfaces. Every control here and every WebMCP tool call hit the same reducer — the trace tags each change [HUMAN] or [AGENT].
Control Deck
Agent Simulator
This box invokes the exact same execute() handlers a real WebMCP agent calls — same schemas, same store reducer, same scene. Nothing here is a mock, so what you run is the shipped code path.
Parts
Click a part to select its assembly. Then, holding Ctrl (or Cmd) over any part:
- click — run what it does. Doors swing, the hood lifts, anything else pops off and back on.
- drag — pull that part out along its own axis, away from the car.
- scroll — resize it.
Double-click a part to fly the camera to it, and empty space to fly back.
109 parts, down to individual handles, wipers and brake discs.
Hand Control
Point at the car with your hand, then pinch your thumb and index finger together to hold a part. Tracking runs in this browser tab - the video is never uploaded or recorded.
- point - a cursor follows the gap between your thumb and index finger, and the part under it glows.
- pinch a door or the hood - hold it and move your hand: it swings on its hinge with you.
- pinch any other part - a window, a wheel, a mirror: carry it, then open your fingers to let go.
- quick pinch - opens or shuts, pops a part off, or puts it back, like Ctrl-click.
- pinch empty space - move your hand to turn the view around the car.
- both hands in empty space - spread to explode, move together to carry the car, twist to turn it. The trace logs each result as [HAND].
Photo to 3D
Depth is estimated by a model running in this browser tab. Your image is never uploaded anywhere. Objects only — people and animals are declined.