AI AI Toolkit
Taking Off TodayMIT

Your Blender Scene, Now With an AI Copilot That Actually Does the Work

⭐ 2.5K Stars

What This Project Actually Does

Think of it as an AI copilot that lives inside Blender and genuinely touches your scene. You describe what you want in plain language — something like "put the product in the center, add a soft top light, and swap the background to light gray" — and it reads the scene, creates objects, assigns materials, arranges layouts, and even runs multi-step workflows from start to finish. You stay in Blender the whole time, watching a viewport preview and confirming the result, instead of alt-tabbing into a terminal. Under the hood it ships an MCP server, so it can talk to cloud models like ChatGPT, Claude and Gemini, or to a local Ollama instance, or to any compatible custom endpoint you point it at. Here's the thing: until now, batch-editing materials or blocking out a product layout meant writing bpy scripts and debugging them. This turns that into a conversation where the assistant does a pass first and waits for your nod before going further. The tagline — "Your Blender scene. Your AI models. One creative workspace." — pretty much sums up the ambition: modeling, rendering and AI all happening at the same desk.

Why It Blew Up Overnight

The repo was created on November 19, 2026, and picked up 530 stars in a little over a day. Almost all of that heat comes from one collision: MCP meeting Blender. MCP has quietly become the de facto standard for AI tools talking to each other, but the 3D world never had a clean way to plug in — you wanted AI to actually move things in your scene, not dump a code snippet for you to paste. Blender users have long been stuck writing bpy scripts for the dull stuff: a batch of materials, a product arrangement, a set of exports. This project wraps that whole capability in a conversational interface and adds a workspace with viewport snapshots and a task board, landing squarely on the "AI walks into the creative tool" trend. MIT licensed, version 1.0.0, Windows x64 only for now — exactly the kind of thing early adopters share the moment it appears.

The Technical Bits That Matter

This isn't a chat wrapper bolted onto a scene editor, and the details show it. First, scene awareness: before acting, it inspects the scene and objects so it knows what's actually there instead of blindly emitting code. Second, the companion window they call Studio is the real center of gravity — chat, scene objects, viewport snapshots, approval buttons and a task board all live together, so you can watch it work step by step and sign off at key points. That matters a lot if you're nervous about an AI rearranging your scene. Third, model choice is wide open: OpenAI, Claude and Gemini in the cloud, Ollama locally, plus custom endpoints, with work and personal connections saved separately so switching projects doesn't mean reconfiguring everything. Fourth, it's a genuine MCP server, not just a client — meaning other MCP clients can turn around and treat Blender itself as a tool. Fifth, multi-agent workflows break a big request into steps, with object creation, material assignment and export prep handled by different agents. It wants Blender 4.2 or newer, currently ships a Windows x64 installer, and the README splits quick-start and connection/agent setup into separate docs.

Who Should Actually Use It

Three groups get the most out of it. 3D creators and indie game developers who have the picture in their head but keep getting tripped up by bpy syntax will find describing faster than coding. Technical artists, who already automate pipelines, get repetitive work turned into an approvable workflow and can spend the saved time on the parts that actually pay. And AI tool developers get a working reference for how MCP can be wired into a professional creative app — the connection layer and scene tools are right there to study. To be fair, if you're doing tightly custom rigging or a large production pipeline, this is still an assistant at version 1.0, not a replacement for the whole job.

Getting It Running

Grab the Windows x64 installer from the GitHub Releases page and have Blender 4.2+ installed. On first launch, configure a model: either drop in a cloud API key or point it at a local Ollama address, and connections are stored separately for work and personal use. Don't start with something ambitious — ask it to "inspect the current scene and list the objects" to get a feel for how it responds, then move on to multi-step workflows once you're comfortable with the approval flow.

How It Stacks Up

  • Against BlenderGPT-style plugins: most of those stop at "here's a bpy snippet, copy it." This adds a viewport preview and an approval workspace, so changes are visible and reversible.
  • Against hand-written bpy scripts: scripts are precise but every change costs writing and debugging. This trades some of that precision for speed, with you still signing off on complex logic.
  • Against Blender's own AI experiments: the official work leans toward low-level capability research, while this one is an out-of-the-box creative workflow that wires in the MCP ecosystem directly.
  • Against general AI assistants: a generic assistant can't see your scene; this one inspects real objects before it touches anything.

Bottom line: it won't beat an expert's hand-written script on depth, but on the distance from idea to picture, it drops the barrier a long way.

🚀

Get Started

Open Source · Commercial Friendly

MIT