
I wanted to know whether an AI could produce a small library of 3D objects for my visuals: abstract shapes, flowers, compositions for flyers. I connected Claude to Blender through the MCP add-on. Here is my verdict.
The setup in brief
Blender, the MCP add-on (a local server to enable and set to start automatically), and the Claude desktop app with the Blender connector. With Claude Code in the terminal, Claude did not receive Blender's replies; the desktop app worked first time. On Windows, remember to allow port 9876 in the firewall.
What the AI does very well
- Changing the colors of a whole scene to match a brand guide, in a few seconds.
- Generating series of objects from a reference style (one image and a style name are enough).
- Fixing things on request: by numbering each object, I could say “redo flowers 1 to 8”.
What still gets stuck
- Completeness: on low effort, Claude forgot some materials. On high effort, everything was fixed.
- Topology: a few hollow meshes, bevels added as modifiers but not applied, parts not merged.
- Rendering: Eevee engine instead of Cycles, flat lighting. I finished by hand with an HDRI from Poly Haven.
- Long conversations: they get expensive, because everything is read again. It is better to start a fresh conversation.
My verdict
For someone who is not a 3D specialist, it is a huge accelerator. But the human eye is still essential for the finishing touches. If you have a serious 3D project, see my services for 3D animation of concepts and prototypes and optimized 3D models.
The full video
A 3D or AI project? Write to me: I reply within 24 to 48 hours on weekdays.












