Skip to main content

Blender MCP

AI INTEGRATION
📖10 min read

Introduction​

The Blender MCP add-on exposes your Blender scene to the Model Context Protocol (MCP), enabling LLMs to query and manipulate objects, materials, and collections through structured tools. It supports both cloud LLMs (fastest setup) and local LLMs (privacy-first).

  • Fastest path: Connect to Claude via the Claude Desktop MCP client (not local; scene data leaves your machine).
  • Local path: Run a local LLM (for example, Ollama) and use the VS Code MCP extension to keep all scene data on-device.

Use MCP when you want repeatable automation, scripted cleanup, or report generation driven by natural language.

Claude (Cloud)

Easiest setup and strongest reasoning. Note: sends scene data to Anthropic servers; avoid for sensitive work.

Local LLM (Ollama)

Run models locally and keep data private. Requires more setup and hardware resources.

Scene Automation

Rename, reorganize, and transform scene elements via natural language instructions executed by the AI.

Installation​

Install the Add-on​

  1. Download the latest release from the GitHub repo: Blender MCP.
  2. In Blender, go to Edit > Preferences > Add-ons.
  3. Click Install from Disk and select the downloaded .zip.
  4. Enable Blender MCP in the add-on list.
  5. Save Preferences.

Prerequisites​

  • Blender 3.0+ (Python-enabled)
  • Python in PATH (for launching the MCP server)
  • For cloud: Claude Desktop + Anthropic API key
  • For local: Ollama installed and a downloaded model

Quick Start: Cloud (Claude)​

Fastest path, but data is not local. Use only for non-sensitive scenes.

  1. Install and enable the add-on in Blender.
  2. Install Claude Desktop and create an API key at console.anthropic.com.
  3. Configure Claude Desktop config.json to register the Blender MCP server: 
    {
      "mcpServers": {
        "blender": {
          "command": "python",
          "args": ["-m", "blender_mcp_server"],
          "env": { "BLENDER_PATH": "/path/to/blender" }
        }
      }
    }
  4. Open your .blend file, launch Claude Desktop, and connect.
  5. Ask Claude to run actions, for example: "List all objects in the 'Buildings' collection and report their bounding boxes."

Privacy warning: Scene structure, object names, and queries are sent to Anthropic servers.

Quick Start: Local (Ollama + VS Code)​

Keeps data on-device. More setup, slower than Claude, but private.

  1. Install Ollama from ollama.ai and pull a model (e.g., ollama pull mistral).
  2. Start the Ollama server (ollama serve).
  3. Install and enable the Blender MCP add-on.
  4. In your workspace, create .mcp.json for VS Code MCP extension: 
    {
      "mcpServers": {
        "blender": {
          "command": "python",
          "args": ["-m", "blender_mcp_server"],
          "env": { "BLENDER_PATH": "/path/to/blender" }
        },
        "ollama": {
          "command": "python",
          "args": ["-m", "ollama_mcp_server"],
          "env": {
            "OLLAMA_BASE_URL": "http://localhost:11434",
            "OLLAMA_MODEL": "mistral"
          }
        }
      }
    }
  5. Open VS Code, install the MCP extension, and start a session.
  6. Run prompts locally, for example: "Move all cameras to Z = 1.6m and point them at the origin."

Core Capabilities​

  • Scene queries: list objects, materials, collections, bounding boxes
  • Object ops: create, rename, delete, parent/unparent, hide/unhide
  • Transforms: set/offset location, rotation, scale with unit-aware inputs
  • Materials: assign existing materials, report missing textures
  • Reports: generate summaries of scene structure and stats

Automation examples:

  • "Rename all objects with prefix old_ to bkp_ and move them to collection Archive."
  • "Find all objects below Z=0, list their names, and move them up by 5 units."
  • "Select all meshes with more than 500k verts and flag them for decimation."

Best Practices​

  • Start with dry-runs: Ask the LLM to describe intended changes before execution.
  • Constrain scope: Specify collections or name patterns to avoid broad edits.
  • Verify after actions: Inspect the Outliner and transforms before saving.
  • Keep backups: Save versions before running large automation steps.
  • Stay aware of privacy: Use Claude only for non-sensitive scenes; prefer local for confidential work.

Summary​

Blender MCP bridges Blender and LLMs via MCP: Claude gives the fastest, most capable cloud setup (but transmits data externally), while Ollama + VS Code keeps everything local with extra setup and hardware needs. Use it to automate scene cleanup, organization, and reporting; always verify AI-made changes before committing them.