Blender MCP is impressive but not that useable (yet)
A user tested Blender MCP, finding it an impressive engineering feat but not yet practical for 3D modeling. Despite multiple iterations and attempts at self-improvement loops, the tool only provided an 80% solution, similar to AI coding. The resulting meshes lacked fundamental structure for decent edge flow, making topology cleanup pointless and requiring a complete rework. The user suggests that better integration methods, akin to harnesses in coding, are needed to make Blender MCP more feasible, as its current state is not good enough for practical use.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 12, 2026, 10:36 UTC
IngestedOffset at this time: UTC+0Sep 12, 2026, 12:01 UTC
- Published
- Sep 12, 2026, 10:36
- Ingested
- Sep 12, 2026, 12:01
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
So after all the hype i wanted to give the blender mcp a try. I know what sub this is and what people in that space are like, but i also believe that giving the AI-Hype train a small reality check from time to time will help to actually improve these things. I do find it an incredible engineering feat to build tools like this in the end, so here a my 2 cents:
I ran multiple iterations, tried to create a self improvment loop, but its like coding with ai. You only get a 80% solution. Unlike coding tho the results arent that easily modifiable. Like if i ask it to write a function for me i need to break it down so much that fits into the requirements of the codebase and maybe write a few lines here and there to simplify things and nudge it into the right direction. Whereas with the mcp i cant see something like this. It did a decent shilouette but cleaning that topology is pointless (this is what i got after i added a prompt that i want quads): Id need to rework the whole mesh anyway, as it simply lacks the fundamental structure to create a decent edge flow. I assume humans will need to find ways to integrate it better. Simmilar as coding got a whole lot better with harnesses, id really wanna see if a simmilar approach could work for blender, but for now its just not that good. Impressive, but not that feasible. Anyone else here that has a fair bit of expierence with 3D Modelling and tried the new astra Model?