Blender MCP: How the Blender Lab Server Works and Its Risks
A guide to the Blender Lab MCP project: its add-on plus server design, about 25 tools, safety risks of running Python, and what we could not verify.
By ShilpBot Team · Published · 4 min read
Blender is the most popular free 3D tool, and an MCP server for it lets an AI assistant look at a scene, run actions and render previews. This guide explains how the Blender Lab project is built, what tools it offers, and the safety points to understand before you use it. Basics first: what MCP is.
Two different projects with similar names
Many Blender MCP projects exist. The one covered here is published on projects.blender.org under Blender's "lab" area. It is separate from popular community projects on GitHub, which are independent and not made by Blender. We could not open the Blender Foundation's own website to confirm the official status, because the page refused to load, so check the Blender Lab page for the current wording.
How it is built
The project's readme describes a lightweight MCP server for Blender with two parts:
- A Blender add-on (an extension) that runs inside Blender.
- An MCP server process that the AI client launches. It talks to the add-on over a TCP socket.
The readme says the add-on must be installed and enabled for any of the MCP tools to work. Its preferences let you set the host, port and optional auto-start.
What tools it offers
The tool list in the project documentation has about 25 tools. Examples include:
- Running Python code in the connected Blender session.
- Running Python in a background Blender process, for working on a file without the interface.
- Summarising objects in a scene and file contents.
- Searching Blender's Python API documentation and the manual.
- Rendering the viewport or a thumbnail to a file, and taking screenshots.
Many tools have a version for command-line use that opens a file in background Blender.
The safety point you must understand
A tool that runs Python code can do anything a Python script can do on your computer, including deleting files. The readme we read does not describe tokens, passwords or tool-group limits, and does not say whether the socket accepts only local connections.
- Use it only on your own computer, and keep the host set to local.
- Work on a copy of your .blend file.
- Do not connect it to an assistant or service you do not trust.
- Watch what code the assistant runs before you accept results.
Limits noted in the project
The readme says deferred responses are supported only by the interactive add-on server, and that background mode requires requests to complete synchronously. The add-on is described as intended to be fairly minimal.
What we could not confirm
The readme does not state a release status, a minimum Blender version or a release date. Search summaries mentioned Blender 5.1 or newer for the official add-on, but we could not verify that on a page we could open, so confirm the requirement before you install.
Good first experiments
- Ask the assistant to summarise the objects in a simple scene.
- Ask it to search the Python API for how to add a modifier, then do it yourself.
- Ask for a thumbnail render and compare it with your own.
Sources
- Blender Lab: Blender MCP project
- Project readme and tool list on projects.blender.org (read 16 Nov 2026).
Frequently asked questions
Is there an official Blender MCP?
Blender Lab publishes a Blender MCP project on projects.blender.org. We could not open Blender's main website to confirm its wording, so check the Lab page. Other popular Blender MCP projects are independent.
Is Blender MCP safe?
It can run Python code in Blender, which can do anything a script can do. Use it only on your own computer, on a copy of your file, with assistants you trust.
What can Blender MCP do?
Its documentation lists about 25 tools, including running Python, summarising scene objects, searching the API and manual, and rendering viewport previews.
