Codex Blender Mcp Demo

Codex Blender MCP Demo

An open, reproducible set of Codex + Blender demos showing how Codex can drive Blender through Python and MCP-style tool execution.

The project now contains two showcase workflows:

• Automatic scene building: Codex creates a complete technical concept

scene from scratch, including geometry, materials, lights, camera, labels, .blend output, a hero render, and a GIF preview.

• Existing asset decomposition: Codex opens a Blender scene from

nidorx/matcaps, isolates the female sci-fi bust, segments it into 12 semantic modules, and keyframes a clean disassembly / reassembly animation.

Demo 1: Codex -> MCP -> Blender

This compact demo uses hassledzebra/codex_blender_mcp in background subprocess mode. It generates a stylized technical workbench scene that visualizes a Codex-to-Blender control loop.

What it demonstrates:

• Codex can call a Blender MCP tool chain from Python.
• Blender can be driven in background subprocess mode.
• A single script can create geometry, materials, lights, labels, camera setup,

keyframes, .blend output, and rendered media.

• The generated scene can be inspected again through blender_scene_info.

Demo 2: MatCap Bust Semantic Assembly

This demo uses the repository imagery and preview bust from nidorx/matcaps. Instead of using generic spheres, it works with the actual female sci-fi bust asset.

Codex opens the source Blender scene, isolates the female sci-fi bust, segments the sculpture body into 12 animatable semantic modules, applies MatCap-style materials, and renders a clean disassembly / reassembly animation:

• cranial_shell
• face_mask
• jaw_neck_front
• rear_head_plate
• temple_ear_module
• neck_spine
• upper_chest_shell
• front_bust_pod
• side_rib_panel
• rear_torso_shell
• lower_base_block
• connector_transition

What it demonstrates:

• Codex can inspect existing Blender geometry instead of only creating new

primitives.

• A sculptural mesh can be converted into semantic modules for explanation,

visualization, or assembly-style animation.

• The final animation is rendered cleanly with only the sculpture and separated

modules visible.

Outputs

Core scene-building outputs:

• outputs/codex_blender_case/codex_blender_mcp_case.blend
• outputs/codex_blender_case/codex_blender_mcp_case_animated.blend
• outputs/codex_blender_case/codex_blender_mcp_case.png
• outputs/codex_blender_case/codex_blender_mcp_case_animation.gif
• outputs/codex_blender_case/x_article_codex_blender_direct_control.md

MatCap bust assembly outputs:

• outputs/matcap_bust_component001_assembly/component001_semantic_assembly.blend
• outputs/matcap_bust_component001_assembly/component001_semantic_assembly_hero.png
• outputs/matcap_bust_component001_assembly/component001_semantic_assembly_animation.gif
• outputs/matcap_bust_component001_assembly/component001_semantic_assembly_summary.json

Project thumbnail / cover:

• docs/cover_5x2.png (2000 x 800, 5:2)

Requirements

• Windows, macOS, or Linux
• Blender 5.1 or newer recommended
• Python 3.10+
• Git

The demos were created and verified with Blender 5.1.2 on Windows.

Quick Start

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
$env:BLENDER_PATH = "C:\Program Files\Blender Foundation\Blender 5.1\blender.exe"
.\.venv\Scripts\python.exe .\tools\run_codex_blender_case.py
.\.venv\Scripts\python.exe .\tools\render_codex_blender_animation.py

If Blender is already on your PATH, you can omit BLENDER_PATH after adapting the scripts or setting it to blender.

The MatCap bust demo expects the nidorx/matcaps scene.blend and PNG MatCap assets to exist under assets/. The generated .blend, PNG, GIF, and JSON outputs are included for reference.

Scripts

• tools/run_codex_blender_case.py creates the first generated scene, saves the

.blend, renders the hero PNG, and writes a scene summary.

• tools/render_codex_blender_animation.py opens the generated .blend, adds

clearer keyframes, renders a PNG frame sequence, and compiles a GIF preview.

• tools/preview_component001_semantic_segmentation.py previews the 12-part

semantic segmentation of the continuous component_001 mesh.

• tools/run_component001_semantic_assembly_demo.py creates the final clean

disassembly / reassembly animation.

Safety Note

MCP-driven Blender workflows execute generated Python code in Blender. Run demos in a clean project directory or virtual machine if you are testing untrusted prompts or third-party scripts.

Credits

• Blender: https://www.blender.org/
• Blender MCP reference: https://www.blender.org/lab/mcp-server/
• Codex Blender MCP: https://github.com/hassledzebra/codexblendermcp
• MatCap assets and preview scene: https://github.com/nidorx/matcaps

License

MIT

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: studyself
• Source: studyself/codex-blender-mcp-demo
• License: MIT

Install and usage instructions live in the source repository linked above.