Hey everyone,
I’ve been building this for a while and finally shipped it, so I figured I’d share here since this community is where I do most of my testing.
The problem
If you’ve used any of the AI-Blender integrations floating around (BlenderGPT, the various MCP forks on GitHub), you’ve probably hit the same wall I did: they either require you to set up Claude Desktop separately and fight with config files, or they just generate raw Python scripts that you have to manually run and debug when they break.
Neither felt like a real workflow tool. I wanted something that lived inside Blender, understood my scene without me explaining it every time, and didn’t fall over the moment I asked for something slightly complex.
What I built
BlenderMCP Pro is an addon that puts a full AI chat panel directly in the N-panel. You type what you want, it does it. No external app required — just an Anthropic API key.
A few things that differentiate it from what’s already out there:
It knows your scene automatically. Every message includes a live snapshot — object names, face counts, materials, active object, and render engine. You’re not explaining your scene every single prompt like you have to with raw script-generation tools.
Full undo support. Every AI action pushes a proper undo step. Ctrl+Z works exactly like it does on any manual action. This was non-negotiable for me — I wasn’t going to ship something that could mess up a scene with no way back.
It’s not just primitives and transforms. This is where most of the free tools stop. BlenderMCP Pro can:
- Build complete shader node graphs (not just “make it red” — actual node-level control, connect arbitrary sockets, build glass/emission materials from scratch)
- Build geometry nodes setups (describe a scatter and it wires the full Distribute Points → Instance → Realize pipeline)
- Batch validate and export for UE5/Unity/Godot with polygon budget checks
- Set up studio lighting rigs in one prompt
Persistent chat history. Conversations save automatically, get titled from your first message, and you can favorite/rename/reload them. Survives a Blender restart.
Viewport screenshot input. Toggle a checkbox and Claude sees exactly what you’re looking at — useful for “why does this shading look wrong” type questions instead of trying to describe it in text.
Works with more than just Claude. It’s built on MCP (the protocol Anthropic introduced for connecting AI to tools), so Cursor and Windsurf connect with one click too, auto-configured from the panel.
Where it’s weaker
Being honest, since I’d want the same if I were buying:
- Rigging tools are new and still rough —
create_basic_rigbuilds a real armature with proper bone hierarchy, but I’m still tuning how reliably Claude reaches for it instead of building mesh primitives. If you test it and it does something dumb, that’s a known area I’m actively improving. - It’s not a replacement for actually knowing Blender. It’s an accelerant for the repetitive stuff — setup, cleanup, export, lighting — not a way to skip learning the software.
- Costs API usage. Typical session runs $0.05–0.15. Not free, but you control the spend.
Example workflow
Real session from yesterday, paraphrased:
“Create a biped rig at the origin, then build a simple low-poly mesh and parent it with automatic weights, then assign a metallic blue PBR material”
Four tool calls, about 8 seconds, fully undoable if I didn’t like it.
Specs
- Blender 5.0+
- Windows/macOS/Linux
- No pip dependencies
- $25 (50% off) on Superhive
Happy to answer questions or take feature requests — this is actively developed, and I’m building based on what people actually need, not just what I think is cool.