How I Built a Blender AI Plugin That Generates 3D Models from Text: 3D-Agent

TLDR: I built Cursor for Blender which generates clean wireframes.

Let me start with a confession.
I'm a terrible 3D artist. Like, genuinely bad. Every time I opened Blender, I'd spend three hours trying to model something simple and end up with a geometry crime that looked like it was designed by someone who has never seen a real object before.

But here's the thing. I needed 3D assets for my projects. Game prototypes, architectural mockups, product visualizations. And hiring a 3D artist for every little thing, specially to send proposals and to make drafts, was not in the budget.

So I did what any developer with too much free time and not enough common sense would do.

I built an AI that does it for me.

Actually I googled first for Blender AI plugin and Blender AI tools and only found Blender-MCP which did not meet my expectations.

The "Aha" Moment

It started when I was playing around with Claude and the Model Context Protocol (MCP). If you haven't heard of MCP, it's basically a way to let AI models interact with external tools and services. Pretty cool stuff.

I thought: what if I could connect Claude to Blender?

What if I could just type "spiral staircase" and have it actually appear in my scene? Like an AI 3D model generator?

Turns out, you can. And it works way better than I expected.

Meet 3D-Agent

3D-Agent is the result of that experiment. It's a Blender addon (written in Python, obviously) that lets you generate 3D models from text descriptions.

You type what you want. The AI figures out how to build it. Blender renders it. Done.

No more dragging vertices around for hours. No more watching YouTube tutorials at 2am trying to understand subdivision surfaces. Just describe your vision and let the machine do the heavy lifting.

How It Actually Works Under the Hood

Since this is producthunt and you probably want the technical details:

The Blender Side

The addon hooks into Blender's Python API. When you submit a prompt, it sends the request to the AI, which returns a series of Blender operations. These get executed sequentially to build your model.

Think of it like a very sophisticated macro system, but instead of recording your clicks, an AI is generating the commands based on your natural language input.

The AI Side

The plugin works with Claude through MCP integration. The model has been given context about Blender's capabilities, common 3D modeling patterns, and best practices for clean topology.

When you say "make a medieval tower," it doesn't just create a cylinder and call it a day. It actually understands architectural concepts and applies them.

The Output

Models export to all the standard formats: OBJ, FBX, GLB, USDZ. The topology is clean enough to use in production. I've shipped game prototypes with assets generated entirely through prompts.

You can watch me using in it on the following YouTube video:
VIDEO

The whole thing took about 10 minutes. Try doing that manually as a beginner or with other Blender AI tools. I'll wait.

Want to Try It?

If you're a developer who needs 3D assets but would rather write code than push vertices around:

👉 Grab the addon here

It's free to try. Works with Blender 3.0 and up. Takes five minutes to install.

And if you run into issues or have feature requests, come hang out in the Discord. I'm pretty active there and always looking for feedback.

What's Next?

I'm constantly adding new features based on what users actually need. Right now I'm working on:

• Better batch generation for creating multiple assets at once
  

• Improved topology optimization for game engines
  

• More export options

If there's something you'd love to see, drop a comment. I actually read them!

Thank you all!