Vi3W
Skip to content

Guides

Can ChatGPT control Blender? The three routes that actually work

Can ChatGPT control Blender? Three working routes: Python scripts, Codex with Blender MCP, and ChatGPT custom MCP apps, plus their limits.

6 min read

A glowing cable plug connecting a glass chat bubble to a small 3D scene on a turntable
Illustration. ChatGPT reaches Blender only through a connector: a script you run, a local agent, or an MCP server.

Quick answer

Can ChatGPT control Blender?

Not by default: the chat window has no way into a desktop app. There are three working routes. Paste ChatGPT's Python into Blender's Scripting tab. Run Codex locally with a Blender MCP server. Or connect a Blender MCP server to ChatGPT as a custom app through a Secure MCP Tunnel, which needs a Business or Enterprise/Edu workspace.

Why the answer isn't a simple yes

ChatGPT runs in OpenAI's cloud. Blender runs on your computer. For ChatGPT to move a vertex, something has to carry its instruction across that gap and into Blender's Python API. There are three ways to do that today. They differ a lot in setup effort, plan requirements and how much you have to trust the AI with your files.

Here they are, from least to most hands-off.

Route 1: ChatGPT writes the script, you run it

This works on every ChatGPT plan, including Free, and it's still the route most people should start with.

  1. In Blender, switch to the Scripting workspace. You get a Text Editor, a Python Console and an Info log showing the Python behind your recent clicks.
  2. Ask ChatGPT for a script and tell it your Blender version: "Blender 5.3. Write a bpy script that…"
  3. Paste it into a new text block and press Run Script.
  4. If it errors, copy the full error from the system console (Window ‣ Toggle System Console on Windows; start Blender from a terminal on macOS and Linux) and paste it back.

Prompts that tend to work first time:

  • "Rename every selected object to Crate_01, Crate_02 and so on, in X order."
  • "Add a Bevel modifier, width 2 cm, 3 segments, to every selected mesh."
  • "Make a low-poly fence: 12 posts 2 m apart along X, with two rails between each pair."
  • "Export each selected object as a separate GLB into a folder called exports next to the .blend."

The upside is control. You read the code before it runs, and nothing touches Blender without you pressing the button. The downside is the copy-paste loop. ChatGPT can't see your scene, so for anything visual you describe it or paste a screenshot.

Tip: the Info editor logs the Python for many UI actions. Doing something once by hand and pasting that line into ChatGPT ("make this run on every selected object") gets you better scripts than describing the action in words.

Route 2: Codex plus a Blender MCP server

This is the practical way to let OpenAI's models drive Blender live. Codex runs on your machine, so it can reach a local MCP server, which ChatGPT on the web can't.

There are two Blender MCP servers:

  • Official Blender MCP from Blender Lab. An add-on plus a blender-mcp server that your client launches over stdio. It relays to Blender over a local TCP connection, and its tools include running Blender Python, summarising the scene and taking screenshots.
  • MCP for Blender, the community project by Siddharth Ahuja. You launch it with uvx mcp-for-blender. It adds Poly Haven assets and optional Sketchfab, Hyper3D Rodin and Hunyuan3D integrations.

Setup in outline:

  1. Install and enable the add-on in Edit ‣ Preferences ‣ Add-ons.
  2. Register the server command in Codex's MCP configuration: blender-mcp for the official server, or uvx mcp-for-blender.
  3. Start the server inside Blender. The community add-on uses the N panel ‣ MCP for Blender ‣ Start MCP Server.
  4. Start Codex in your project folder and ask something read-only: "List the objects in the open Blender scene."

The client-by-client details are in Blender MCP with Codex, Cursor and Gemini. If the connection drops, Blender MCP not working covers the usual causes.

A glass chat bubble connected to a small 3D scene by a beam of light
Illustration. The MCP server turns the model's tool calls into Python that runs inside Blender.

Route 3: ChatGPT itself, through a custom MCP app

ChatGPT on the web can use MCP servers, but with two catches for Blender.

Catch one: plan. OpenAI's help centre says custom MCP server connections with full MCP support are for ChatGPT Business and Enterprise/Edu on the web, not mobile. Pro users can connect custom servers with read and fetch permissions only, which means no write actions, so no editing your scene.

Catch two: ChatGPT only talks to remote servers. It connects from OpenAI's infrastructure to a public HTTPS endpoint. A Blender MCP server on localhost isn't reachable. OpenAI's answer is a Secure MCP Tunnel, which links a server on a private network or developer machine to ChatGPT without exposing it to the public internet.

If you clear both, the flow is: Plugins ‣ plus button ‣ Add custom MCP server, choose Tunnel under Connection, set authentication, accept the risk warning, then Create as a plugin. In a chat, type @ and pick the plugin. ChatGPT asks for confirmation before write actions. Full steps: connect an MCP server to ChatGPT.

Honestly, for a solo artist this is the most effort for the least gain. Route 2 gets you the same control with fewer moving parts.

Which route should you pick?

  • You want help learning Blender, or a one-off script: Route 1.
  • You want an AI to rearrange scenes, batch-fix materials or test add-ons while you watch: Route 2.
  • Your studio already runs on ChatGPT Enterprise and IT wants one governed connection: Route 3, with your admin.

What ChatGPT can't do in Blender

  • Sculpt a good character. Through Python it builds from primitives, modifiers and geometry nodes. Ask for a dragon and you get something assembled from spheres. Organic shapes come out better from an image-to-3D generator, imported into Blender afterwards.
  • See without being shown. Over MCP it can request a viewport screenshot. In plain chat it only knows what you tell it.
  • Stay in the lines with Python. Every Blender MCP server here can run arbitrary Python. Save your .blend first. The official project suggests a VM or isolated machine, and the community add-on's socket has no authentication, so keep it on localhost.
  • Keep up with very new API changes. Tell it your Blender version, and paste the error when a script uses a renamed operator.

Ten prompts worth keeping

  • "List every object with its triangle count, sorted highest first."
  • "Find materials with missing image files."
  • "Apply scale and rotation on everything selected."
  • "Set every selected object's origin to the bottom centre."
  • "Add a Weighted Normal modifier to every selected mesh."
  • "Join all objects whose name starts with Wall_."
  • "Make a 5 x 5 grid of the selected object, 2 m apart."
  • "Rename everything in the Props collection with a Prop_ prefix."
  • "Export each selected object as GLB into an exports folder."
  • "Take a viewport screenshot." (MCP only)

A sensible split of work

Let ChatGPT do the code-shaped jobs: naming, arranging, modifiers, exports, add-on scaffolding (write a Blender add-on with ChatGPT). Get the detailed props from a generator. Then ask the assistant to import, scale and place them. If you're choosing between assistants for this, see ChatGPT vs Claude for Blender.

Bring in models ChatGPT can't build

Vi3W makes the meshes that Python can't: text, sketch or image to 3D, then Add bones, Make it move, New paint job and Clean shape, and GLB or FBX out for Blender. With Vi3W's MCP server you can do it from the same chat. Try Vi3W.

Frequently asked questions

can i connect chatgpt to blender

Yes, but not with a plain chat. The simple route is to copy ChatGPT's Python into Blender's Text Editor and run it. For live control, use Codex on your computer with a Blender MCP server. Connecting ChatGPT itself needs a workspace plan with full MCP and a tunnel, because Blender runs on your machine and ChatGPT only reaches remote servers.

does chatgpt have a blender plugin

There's no official ChatGPT add-on for Blender. What exists is Blender MCP: the official Blender Lab server and the community MCP for Blender add-on. Both work with any MCP client, including Codex, and ChatGPT workspaces can connect one as a custom app.

does chatgpt work with blender

It works well as a Blender tutor and Python writer on any plan. It can also drive Blender directly through MCP, mostly via Codex. Expect it to be strongest at code-shaped tasks (renaming, arranging, modifiers, batch exports) and weakest at organic modelling.

Last reviewed