Blender MCP is impressive but not that useable (yet)

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?