feat(mcp): detect floorplan openings stairs and furniture

packages/mcp/examples/photo-to-scene.md

@@ -52,7 +52,10 @@ Optional knobs:
    `sampling_response_invalid` MCP errors.
 3. A fresh `SceneGraph` is built using the core schema factories: a
    `site` → `building` → `level 0` skeleton, then one `WallNode` per
-   vision wall and one `ZoneNode` per vision room. Each node is
+   vision wall and one `ZoneNode` per vision room. Detected doors and windows
+   are attached to the nearest wall, stairs get a stair plus segment node, and
+   detected furniture is matched to the built-in catalog and placed with
+   clearance checks. Each node is
    re-parsed with `AnyNode.safeParse`; invalid ones are dropped with a
    warning appended to `notes`.
 4. `bridge.setScene(...)` swaps the live scene so any follow-up MCP call
@@ -69,6 +72,10 @@ Optional knobs:
   "url": "/scene/scene_01hx8a...",
   "walls": 4,
   "rooms": 1,
+  "doors": 2,
+  "windows": 4,
+  "stairs": 1,
+  "furniture": 6,
   "confidence": 0.82
 }
 ```
@@ -116,4 +123,6 @@ wall, and issues `cut_opening` — no extra wiring needed.
 
 - `photo_to_scene` is a one-shot primitive: one call, one scene.
 - Vision confidence is surfaced so the agent can warn the user.
-- v0.1 covers walls + zones; doors, windows, items are follow-up tools.
+- The one-shot builder returns walls, rooms, openings, stairs, and matched
+  catalog furniture. Items that cannot be matched or placed are reported in
+  `notes` rather than silently dropped.

packages/mcp/src/tools/photo-to-scene/photo-to-scene.test.ts

@@ -70,6 +70,13 @@ const VALID_VISION_JSON = {
       approximateAreaSqM: 20,
     },
   ],
+  doors: [{ position: [2.5, 0], widthM: 0.9, swingDirection: 'inward' }],
+  windows: [{ position: [5, 2], widthM: 1.2, heightM: 1.2, sillHeightM: 0.9 }],
+  stairs: [{ position: [1, 1], widthM: 1, runLengthM: 2.5, rotationDeg: 90 }],
+  furniture: [
+    { type: 'sofa', position: [2.5, 2.8], rotationDeg: 0, widthM: 2.2, depthM: 0.9, confidence: 0.9 },
+    { type: 'coffee-table', position: [2.5, 1], rotationDeg: 0, confidence: 0.86 },
+  ],
   approximateDimensions: { widthM: 5, depthM: 4 },
   confidence: 0.82,
 }
@@ -103,6 +110,10 @@ describe('photo_to_scene', () => {
       url?: string
       walls: number
       rooms: number
+      doors: number
+      windows: number
+      stairs: number
+      furniture: number
       confidence: number
     }
     expect(structured.walls).toBe(4)
@@ -124,6 +135,15 @@ describe('photo_to_scene', () => {
     const zones = allNodes.filter((n) => n.type === 'zone')
     expect(walls.length).toBe(4)
     expect(zones.length).toBe(1)
+    expect(allNodes.filter((n) => n.type === 'door').length).toBe(1)
+    expect(allNodes.filter((n) => n.type === 'window').length).toBe(1)
+    expect(allNodes.filter((n) => n.type === 'stair').length).toBe(1)
+    expect(allNodes.filter((n) => n.type === 'stair-segment').length).toBe(1)
+    expect(allNodes.filter((n) => n.type === 'item').length).toBe(2)
+    expect(structured.doors).toBe(1)
+    expect(structured.windows).toBe(1)
+    expect(structured.stairs).toBe(1)
+    expect(structured.furniture).toBe(2)
 
     // Scene was persisted in the store.
     const saved = await store.load(structured.sceneId!)