Blender MCP
Concept to placeholder 3D asset in 20 minutes. Artists unblocked.
The Problem
An indie game studio making a 3D action RPG had six artists and one 3D generalist. The generalist was the bottleneck. When a concept artist needed a placeholder character model — “a dwarf warrior with heavy plate armour, stocky proportions, carrying a two-handed axe” — they had to wait for the generalist’s queue to clear. That queue was two weeks long.
Concept art was piling up. Environmental designs were blocked waiting for geometry. Animation couldn’t start without meshes. The studio didn’t have the budget to hire a second generalist.
What We Built
Blender MCP extended with two additions specific to the studio’s pipeline:
- CSM.ai mesh retrieval — text descriptions are vectorised and searched against CSM.ai’s 3D asset library; the closest-matching mesh is imported into Blender as a starting point
- Mixamo animation retargeting — once a character mesh is in Blender, a single tool call rigs it, uploads to Mixamo, retrieves an animation, and applies it back to the scene
The pipeline: Claude describes scene → FastMCP server → TCP socket → Blender addon → bpy API.
How It Works
The Blender addon runs a persistent TCP socket server inside the application. Commands arrive as JSON and execute against the bpy API — Blender’s Python scripting interface:
# Blender addon: command handler def execute_command(cmd: dict) -> dict: action = cmd["action"] if action == "import_csm_mesh": # Query CSM.ai for closest mesh to text description mesh_url = csm_search(cmd["description"]) bpy.ops.import_scene.gltf(filepath=download_temp(mesh_url)) obj = bpy.context.selected_objects[0] return {"status": "ok", "object_name": obj.name} elif action == "apply_mixamo_rig": obj_name = cmd["object_name"] obj = bpy.data.objects[obj_name] # Export, rig via Mixamo API, reimport export_path = export_for_mixamo(obj) rigged_path = mixamo_auto_rig(export_path) bpy.ops.import_scene.fbx(filepath=rigged_path) return {"status": "ok", "rigged_object": rigged_path} elif action == "set_material": obj = bpy.data.objects[cmd["object_name"]] mat = bpy.data.materials.new(name=cmd["material_name"]) mat.use_nodes = True # Set base colour from hex colour = hex_to_rgba(cmd["colour"]) mat.node_tree.nodes["Principled BSDF"].inputs[0].default_value = colour obj.data.materials.append(mat) return {"status": "ok"} The conversation loop lets Claude iterate: “make the armour more ornate” → material parameter adjustments; “he’s too tall” → scale transform on specific bones.
The Outcome
Concept artists describe what they need in plain language. Twenty minutes later they have an animated placeholder in their scene. The generalist’s queue cleared within two weeks as the high-volume, low-complexity requests moved to the AI pipeline.
The generalist now focuses exclusively on hero assets — characters and objects that need custom topology, hand-painted textures, and technical artistry. Quality improved while throughput increased.