Skip to content
Tresca

Blender MCP (Blender Lab)

MCP serverOfficial

Blender's own small MCP server and add-on: runs Python in a live or background Blender and summarises .blend files. By Blender Lab.

At a glance

Type
MCP server
Works with
Blender / Bonsai (IFC)
How it connects
Drives a running copy of the tool
Access
Can modify models
Tool versions
Blender 5.1 or newer (add-on 1.0.3 on the Blender Lab page)
Operating systems
Not stated in its documentation
Needs a Blender / Bonsai (IFC) licence
No
Transport
Local (stdio)
AI clients documented
llama.cpp
Licence
GPL-3.0-or-later
Language
Python

What it can do

  • Runs Python sent by the agent inside an open Blender session, or in a background Blender process that opens a .blend file you name.
  • Summarises a blend file without writing code, including data-block counts, missing external files, linked libraries and a guess at what the file is used for.
  • Looks up the Blender Python API reference and manual excerpts bundled with the server, so the agent can check bpy names before it writes code.
  • Takes screenshots of the Blender window or a single area and describes the window layout and selection.

Limitations

  • Needs Blender 5.1 or newer; older releases can't load the add-on.
  • Blender Lab warns that agent-written code runs without safeguards and could delete data or send it elsewhere, and suggests a virtual machine or a system with no sensitive data.
  • The setup page names only llama.cpp; other MCP clients should work but have no instructions there.
  • Deliberately minimal, with no BIM, IFC or Bonsai-specific tools.

Install

Install the add-on in Blender by dragging the zip onto Blender twice (first to add the Blender Lab repository, then to install), or use Install from Disk. Then install the MCP server either as an MCP Bundle (.mcpb) from the release page, for clients that support it, or from source by following the linked llama.cpp instructions.

Source: www.blender.org/lab/mcp-server/

MCP servers and skills run code on your computer. Review the source before installing.

AI-generated models, drawings and results must be checked by a qualified engineer before use.