> ## Documentation Index
> Fetch the complete documentation index at: https://srk-e37e8aa3.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Spatial Context Protocol — real-time AI execution runtime for one embodied system.

# What is SCP

SCP gives one body to one LLM. The body runs at 60Hz, holds its own decision cache, and only calls the brain when the cache cannot answer. After a few sessions on the same task, the brain stops being called at all.

```bash theme={null}
npm install scp-protocol
```

If you have several bodies to coordinate, install [Plexa](/plexa/introduction) on top.

## How it differs from MCP

|                | MCP                      | SCP                             |
| -------------- | ------------------------ | ------------------------------- |
| Who initiates  | Brain asks, tool answers | Body acts, brain advises        |
| Body           | Passive (waits)          | Active (60Hz tick loop)         |
| Memory         | None                     | Pattern store + adaptive memory |
| Cost over time | Constant                 | Drops as novelty decreases      |

## What SCP adds to MCP

1. **Body class with a tick loop**. `SCPBody` runs at 60Hz, owns its sensors, owns its actuators. The brain is a slow advisor, not the driver.
2. **Pattern store**. Every brain decision is cached against a feature vector. Next time the same situation appears, the body answers locally in 0.1 ms.
3. **Adaptive memory**. When the cache misses but a similar situation has been seen, a similarity-scored layer answers without a brain call.

## Quick start

```javascript theme={null}
const { SCPBody, PatternStore, AdaptiveMemory } = require("scp-protocol")
const { OllamaBridge } = require("scp-protocol/bridges/ollama")

class Patrol extends SCPBody {
  static bodyName = "patrol"
  static tools = {
    halt: { description: "stop", parameters: {} },
    advance: {
      description: "move forward",
      parameters: { speed: { type: "number", min: 0, max: 1, required: true } },
    },
  }
  async halt() { /* drive hardware */ }
  async advance({ speed }) { /* drive hardware */ }
}

const body = new Patrol({
  patternStore: new PatternStore({ featureExtractor: (e) => ({ kind: e.kind }) }),
  adaptiveMemory: new AdaptiveMemory({ threshold: 0.8 }),
  brain: new OllamaBridge({ model: "llama3.2" }),
})
```

That is the whole contract a body owner has to write. Sensor reading, brain calling, and outcome reporting happen inside the base class.

## Examples

See [/examples/cart-pole](/examples/cart-pole) for a complete walkthrough that goes from `npm install` to a balanced inverted pendulum in under ten minutes.