Skip to content
Agentmart
Submit a tool

blender-mcp

Create, iterate, and export 3D scenes in Blender via MCP with visual feedback loops.

Skill by uset82

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/uset82-blender-mcp

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Create, iterate, and export 3D scenes in Blender via MCP with visual feedback loops. Use for product visualization, interior concepts, motion design, hero renders, or when StackSelection includes the blender-mcp integration. Requires a running Blender instance with the MCP addon enabled.

Use cases: design, data, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/blender-mcp

What's inside

Skill name
blender-mcp
Description
Create, iterate, and export 3D scenes in Blender via MCP with visual feedback loops. Use for product visualization, interior concepts, motion design, hero renders, or when StackSelection includes the blender-mcp integration. Requires a running Blender instance with the MCP addon enabled.
License
MIT
Instructions
134 lines in SKILL.md · view on GitHub

Files (5)

  • SKILL.md
  • claude.yaml · agents/
  • openai.yaml · agents/
  • codex-gpt-blender-mcp.md · docs/
  • integration-img2threejs.md · docs/

SKILL.md (first lines)

# blender-mcp — Blender Model Context Protocol Integration

Build, render, inspect, and fix 3D scenes in Blender through MCP tool calls with a visual
self-correction loop. The agent describes the scene, Blender creates it, the agent sees the
render, and corrections continue until the result matches the original intent.

**Primary path in WebDesigner: Codex / GPT-5.4** (vision + coding + tool use), routed via
`skillOverrides.blender-mcp` in `.antigravity/runtime/routing-policy.json`. Still host-portable:
Claude Code, Codex IDE, OpenCode, or WebDesigner can run the same MCP loop. Wherever this doc
says "viewport screenshot" or "agent vision", use whatever visual feedback the MCP server provides.

**Narrative / product essay:** [docs/codex-gpt-blender-mcp.md](./docs/codex-gpt-blender-mcp.md)

## WebDesigner Contract

- **Stage**: `design` (3D concept development), `build` (scene creation and iteration)
- **Reads**: `TaskIntent`, `StackSelection`, design brief, reference images, generated workspace
- **Emits artifacts**: `blender-scene-spec`, `blender-render`, `blender-review-log`, `exported-asset`
- **MCP dependency**: `blender` MCP server (via `uvx blender-mcp` or local addon)
- **Activation**: `StackSelection.integrations` contains `blender-mcp`, or `TaskIntent.constraints.requiresBlenderMCP` is true, or the user explicitly asks for Blender scene creation

### WebDesigner Rules

- Require a running Blender instance with the MCP addon connected on `localhost:9876`.
- Use only project-scoped MCP tools: `get_scene_info`, `get_object_info`, `get_screenshot`, `execute_blender_code`.
- Treat `execute_blender_code` as arbitrary local code execution; respect approval prompts when configured.
- Keep PolyHaven, Sketchfab, Hyper3D, Hunyuan, and other network-backed integrations disabled per project policy.
- Export GLB/PNG assets to `.codex-avatar/exports/blender/` or the generated workspace's asset directory.
- Work on explicit copies; never overwrite user-selected source scenes.
- Record render screenshots, comparison sheets, and export paths in the `ArtifactManifest`.

### WebDesigner Process

1. Confirm `integrations` contains `blender-mcp` or user explicitly requested Blender scene work.
2. Verify Blender MCP connection via `get_scene_info`.
3. Parse the user description into a scene specification (objects, materials, lighting, camera).
4. Follow the iterative build loop: create → render → inspect → fix.
5. Export final assets (GLB, PNG, optional blend) to the generated workspace.
6. Emit `blender-scene-spec`, `blender-render`, and `blender-review-log` artifacts.

## When To Use

- Product visualization: bottles, electronics, furniture, fictional devices
- Interior concepts: room layouts, furniture arrangements, lighting studies
- Motion design: procedural forms, extruded typography, geometric loops
- Hero renders: landing page backgrounds, marketing materials
- Asset creation: 3D props for web/game projects, WebGL/Three.js source assets
- Prototyping: quick 3D mockups before committing to detailed procedural code

## Core Promise

Build from description in stages — never attempt a complete scene in one prompt:

1. **Composition first**: Block out main objects, camera angle, basic layout.
2. **Materials and lighting**: Apply initial materials, establish lighting rig.
3. **Render and inspect**: Capture viewport/render, identify visible mismatches.
4. **Targeted corrections**: Adjust specific settings (camera, lighting, materials) based on inspection.
5. **Iterate**: Repeat render-inspect-fix until the result matches intent.

State explicitly when output is approximate or stylized. A single description cannot specify
every detail — ask for clarification instead of making arbitrary choices.

## Required Inputs

- Scene description (objects, mood, purpose, camera angle preferences)
- Output format: product shot, interior render, animated loop, exportable asset
- Quality level: quick prototype, polished render, production asset

## The Loop (MCP tools do execution; agent vision does judgment)

### Available MCP Tools

| Tool | Purpose |
|------|---------|
| `mcp__blender__get_scene_info` | Inspect current scene state |
| `mcp__blender__get_object_info` | Get details about specific objects |
| `mcp__blender__get_screenshot` | Capture viewport for visual feedback |
| `mcp__blender__execute_blender_code` | Run Python code against Blender's bpy API |

### Iteration Process

1. **Scene setup**: Create base geometry with `execute_blender_code`:
   ```python
   import bpy
   bpy.ops.mesh.primitive_cylinder_add(radius=0.7, depth=1.6)
   bpy.ops.object.light_add(type="AREA", location=(4, -4, 6))
   bpy.context.object.data.energy = 800
   bpy.ops.object.camera_add(location=(5, -6, 3))
   bpy.context.scene.camera = bpy.context.object
   ```

2. **Render capture**: Use `get_screenshot` to see current state.

3. **Visual inspection**: Analyze the screenshot with agent vision:
   - Is the subject framed correctly?
   - Are materials reading as intended?
   - Is lighting flattering or problematic?
   - Does composition match the description?

4. **Correction**: Issue targeted fixes based on inspection:
   - Camera too close? Increase distance, decrease focal length.
   - Harsh shadows? Soften light, adjust angle.
   - Material too shiny? Reduce roughness, check reflection.
   - Object hidden? Adjust position, check visibility.

5. **Record review**: Log the iteration with scores and decision:
   - `continue`: Result matches intent, proceed to next stage.
   - `refine-scene`: Adjustments needed, continue iteration.
   - `request-input`: Need clarification from user.
   - `stop`: Cannot achieve requested result, explain limitation.

## Gates (do not skip)

- **Connection check**: Verify Blender MCP is reachable before scene work.
- **Screenshot feedback**: Corrections must be based on actual render output, not assumptions.
- **User approval**: Major direction changes require confirmation.
- **Export validation**: Verify exported assets are valid (GLB header, PNG signature, file size).

## Limitations

From uset82/webdesigner, MIT.

More from uset82/webdesigner

Signals

Stars
0
Forks
0
Open issues
1
Last commit (this folder)
Jul 22, 2026 (3 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from uset82

Cite this page

blender-mcp skill. Agent Mart. https://agentmart.dev/skills/uset82-blender-mcp (last updated 2026-10-24)