# Token-Efficient Workflow — threejs-devtools-mcp

## The problem

A full `scene_tree` JSON can be **60KB+** — that's a huge chunk of context wasted on data you don't need.

## The solution

`scene_tree` returns **compact text by default** (~2KB, 97% smaller). Use targeted tools for details.

## Recommended workflow

```
1. scene_tree              → compact overview (~2KB)
2. Find object by name
3. object_details(name)    → specifics for one object
4. material_details(name)  → material properties
5. set_* tools             → make changes
```

## Examples

### "What's in the scene?"

```
→ scene_tree
← sky [Group]
    skyGradient [Mesh]
    sunDisk [Mesh]
    sunLight [DirectionalLight] intensity=2
  environment [Group]
    env_0 [InstancedMesh] instances=42
    env_1 [InstancedMesh] instances=18
  road [Group]
    road_0 [Mesh]
    road_1 [Mesh]
    ...
```

### "What material does the road use?"

```
→ material_details(name="road_0")
← { type: "MeshStandardMaterial", color: "#8B7355", map: "road_diffuse.png", ... }
```

### "I need full JSON with positions"

```
→ scene_tree(compact=false, depth=2)
← { name: "Scene", type: "Scene", children: [ { name: "sky", position: [0, 10, 0], ... } ] }
```

## When to use `compact=false`

- You need exact positions/rotations of multiple objects
- You're debugging a layout or transform issue
- You need geometry/vertex counts for the whole scene
- Otherwise, always use the default compact mode