SCP connects any LLM to any body without retraining. The muscle layer runs at 60fps. The brain wakes only when the muscle cannot handle something. A pattern store learns from every brain decision and replays them locally over time. Result: brain calls dropped from 27 to 0 across 5 adapters including MuJoCo real physics simulation. Pole balanced 64 loops. Brain went silent by loop 17. npm install scp-protocol

HiBird — Smarter communication for growing businesses.

Smarter communication for growing businesses.

I built this accidentally. I was training AnimTOON-3B, an open model that rigs SVG characters into Lottie animations. Tried to extend it to 3D rigging. Hit a wall. The model was trying to understand the world AND control a body simultaneously. No model is good at both. That question led to SCP. Three additions to MCP: 1. Body describes itself on connect. Swap body, swap JSON. 2. Body pushes events up without being asked. 3. Body runs at 60fps. Brain drops in when needed. In MCP, the brain asks. In SCP, the muscle asks. Five adapters proven. Same LLM. Same protocol. Missile defense, self-driving car, highway, MuJoCo cart-pole, MuJoCo ant. Brain calls: 27 in loop 1. Zero by loop 17. Pole balanced 64 loops without falling. npm install scp-protocol Happy to answer anything.

What use case would benefit most from near-zero API costs after the learning phase?

HiBird — Smarter communication for growing businesses.

Smarter communication for growing businesses.

SCP - Spatial Context Protocol

Any LLM. Anybody. Brain learns. Then goes silent.

Any LLM. Anybody. Brain learns. Then goes silent.

Forum Threads

SCP reduces brain calls from 27 to 0 over sessions

Forum Threads

SCP reduces brain calls from 27 to 0 over sessions