Threlte

npm version npm downloads License: MIT MCP Compatible Built with Claude

An MCP (Model Context Protocol) server that enables AI agents to inspect and manipulate Three.js/Threlte scenes in real-time.

✅ Compatible With

• Claude Desktop - Anthropic's desktop app
• Antigravity - Google's AI IDE
• Claude Code - CLI tool for Claude
• Cursor - AI-powered code editor
• Windsurf - Codeium's AI IDE
• Continue - VS Code AI extension
• Any MCP-compatible client

See detailed compatibility guide →

Features

• 🔍 Scene Inspection - View the full 3D scene hierarchy, find objects by name/type
• 🎯 Object Manipulation - Move, rotate, scale, show/hide objects
• 🎨 Materials & Assets - Apply materials, load GLTF models, change environment
• 🧪 Asset Analysis - Inspect GLTF structure and validate performance
• dYZ+ Asset Optimization - Simplify meshes, compress textures, prune unused data
• dY"? Svelte Export - Generate Threlte/Svelte components from GLTF
• dYZ? Camera Presets - Save, load, and animate camera views
• ⚡ Physics Control - Add physics bodies, apply impulses, set gravity
• 🎭 Vibe Presets - Apply mood presets (cozy, spooky, neon, etc.)

Installation

npm install threlte-mcp

Quick Start

1. Install the package

npm install threlte-mcp

2. Configure your IDE (one-time setup)

npx threlte-mcp setup

This auto-detects your IDE (Antigravity, Cursor, Claude Desktop) and creates the MCP config.

3. Add the component to your Threlte app

Requires Svelte 5 for the MCPBridge component.

<script>
  import { MCPBridgeComponent } from 'threlte-mcp/client';
</script>

<!-- Add inside your <Canvas> -->
<MCPBridgeComponent />

That's it! The AI can now inspect and manipulate your scene.

{
  "mcpServers": {
    "threlte-mcp": {
      "command": "npx",
      "args": ["-y", "threlte-mcp"]
    }
  }
}

import { MCPBridge } from 'threlte-mcp/client';
import { useThrelte } from '@threlte/core';

const { scene } = useThrelte();

// Create bridge (auto-connects in dev mode)
const bridge = new MCPBridge(scene, {
  url: 'ws://127.0.0.1:8082',  // default
  autoConnect: true,           // default in dev
  reconnectDelay: 60000,       // 1 minute
});

// In render loop
bridge.update();

// Get scene state
bridge.getSceneState();

// Find objects
bridge.findObjects({ nameContains: 'player' });

Option B: Per-Scene (Simple for prototyping)

<script lang="ts">
  import { onMount, onDestroy } from 'svelte';
  import { useThrelte } from '@threlte/core';

  const { scene } = useThrelte();
  let ws: WebSocket | null = null;
  
  onMount(() => {
    ws = new WebSocket('ws://localhost:8082');
    
    ws.onopen = () => {
      console.log('[MCPBridge] Connected');
      // Send initial scene state
      ws?.send(JSON.stringify({
        type: 'sceneState',
        data: serializeScene(scene)
      }));
    };
    
    ws.onmessage = (event) => {
      const command = JSON.parse(event.data);
      handleCommand(command);
    };
  });
  
  onDestroy(() => {
    ws?.close();
  });
  
  function serializeScene(obj: THREE.Object3D, depth = 0, maxDepth = 3) {
    if (depth > maxDepth) return null;
    return {
      name: obj.name,
      type: obj.type,
      position: obj.position.toArray(),
      rotation: obj.rotation.toArray().slice(0, 3),
      scale: obj.scale.toArray(),
      visible: obj.visible,
      children: obj.children.map(c => serializeScene(c, depth + 1, maxDepth)).filter(Boolean)
    };
  }
  
  function handleCommand(cmd: any) {
    const { action, requestId, ...params } = cmd;
    let result;
    
    switch (action) {
      case 'getFullSceneState':
        result = { data: serializeScene(scene, 0, params.maxDepth || 3) };
        break;
      case 'moveSceneObject':
        const obj = scene.getObjectByName(params.name || params.path);
        if (obj && params.position) {
          obj.position.set(...params.position);
          result = { success: true };
        }
        break;
      // Add more handlers as needed
    }
    
    if (requestId) {
      ws?.send(JSON.stringify({ ...result, requestId }));
    }
  }
</script>

3. Run your app and start using MCP tools

Once both the MCP server and your Threlte app are running, AI agents can:

"Get the scene state"
"Find all objects named 'Player'"  
"Move the 'Camera' to position [0, 5, 10]"
"Apply the 'neon' vibe to the scene"

Available Tools

Scene Inspection

Camera

Hierarchy Management

Physics

Asset Processing

Materials & Assets

Atmosphere

Configuration

WebSocket Connection

The server uses port 8082 by default for WebSocket communication.

Environment Variables

The MCPBridge auto-connects in development mode. For production, you can enable it via environment variable:

# Add to .env file:
VITE_MCP_ENABLED=true

This is useful for:

• Testing in production builds
• Demo environments
• Debugging production issues

Note: Auto-enabled in development mode (npm run dev), no configuration needed.

Development

# Clone the repo
git clone https://github.com/RaulContreras123/threlte-mcp.git
cd threlte-mcp

# Install dependencies
npm install

# Run in development
npm run dev

# Build for production
npm run build

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

MIT © Raul Contreras

Links

• npm Package
• GitHub Repository
• Model Context Protocol
• Threlte Documentation

Built with Claude 🤖