Blend AI MCP Server

blend-ai

The most intuitive and efficient MCP Server for Blender. Control Blender entirely through AI assistants like Claude — create 3D models, set up scenes, animate, render, and more, all through natural language.

This was created via Claude Code using the Haiki model and 20 random reference images. It took 5 minutes:

Key Features

• 108 tools covering every major Blender domain: modeling, materials, lighting, camera, animation, rendering, sculpting, UV mapping, physics, geometry nodes, rigging, curves, collections, file I/O, and viewport control
• No arbitrary code execution — every operation is an explicit, validated, parameterized tool. No exec(), no eval(), no script injection vectors.
• Zero-dependency Blender addon — the addon uses only Python stdlib + bpy. Nothing to pip install inside Blender's bundled Python.
• Thread-safe architecture — background TCP server with queue-based main-thread execution, respecting Blender's single-threaded API constraint.
• MCP resources — browse scene objects, materials, and scene info as structured context.
• Workflow prompts — pre-built prompt templates for common tasks (product shots, character base meshes, scene cleanup, turntable animations).

Quickstart

1. Install the MCP server

git clone https://github.com/jabberwock/blend-ai.git
cd blend-ai
uv pip install -e .

2. Install the Blender addon

1. Open Blender (5.0+)
2. Go to Edit > Preferences > Add-ons > Install...
3. Select the addon/ folder from this repo (or zip it first)
4. Enable "blend-ai" in the addon list

Alternatively, symlink for development:

# macOS
ln -s "$(pwd)/addon" ~/Library/Application\ Support/Blender/5.0/scripts/addons/blend_ai

# Linux
ln -s "$(pwd)/addon" ~/.config/blender/5.0/scripts/addons/blend_ai

# Windows (run as admin)
mklink /D "%APPDATA%\Blender Foundation\Blender\5.0\scripts\addons\blend_ai" "%cd%\addon"

Then enable the addon in Blender preferences.

3. Start the server in Blender

In Blender's 3D Viewport, open the N-panel (press N), find the blend-ai tab, and click Start Server. The addon listens on 127.0.0.1:9876.

4. Connect your AI assistant

This repo includes an `mcp.json` config file you can use directly or copy into your client's configuration.

Claude Code Integration

claude mcp add blend-ai -- uv run --directory /path/to/blend-ai blend-ai

Replace /path/to/blend-ai with the actual path to your clone. Make sure Blender is running with the addon server started before using the tools.

Usage

$ claude

> Create a red metallic sphere on a white plane with three-point lighting

> Add a subdivision surface modifier to the sphere and set it to level 3

> Set up a turntable animation and render it to /tmp/turntable/

Claude Desktop Integration

Add blend-ai to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "blend-ai": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/blend-ai", "blend-ai"]
    }
  }
}

Replace /path/to/blend-ai with the actual path to your clone. Or copy the contents of the bundled `mcp.json` into your config file.

Restart Claude Desktop. The Blender tools will appear in the tool list.

Other MCP Clients

blend-ai is a standard MCP server using stdio transport. Any MCP-compatible client can connect using the `mcp.json` config or by running the server directly:

uv run --directory /path/to/blend-ai blend-ai
# or: python -m blend_ai.server

The server communicates over stdin/stdout using the MCP protocol. It connects to Blender's addon over TCP on 127.0.0.1:9876.

Architecture

AI Assistant <--stdio/MCP--> blend-ai server <--TCP socket--> Blender addon <--bpy--> Blender

• MCP Server (src/blend_ai/): Python process using the mcp SDK. Exposes tools, resources, and prompts over stdio. Validates all inputs before forwarding to Blender.
• Blender Addon (addon/): Runs a TCP socket server inside Blender on a background thread. Commands are queued and executed on the main thread via bpy.app.timers to respect Blender's threading model.
• Protocol: Length-prefixed JSON messages over TCP. Each message is a 4-byte big-endian length header followed by a UTF-8 JSON payload.

Tool Domains