Blender MCP · from the Reel

Control Blender with AI: the full setup

Blender MCP is a free community plugin that connects any AI assistant to Blender, so you model, texture and light a scene by describing it. The Godzilla in the Reel took about nine minutes and no Blender knowledge.

Watch the tutorial first, then work through the setup alongside it: install uv, install the add-on, point your AI app at Blender, and connect from inside Blender.

A practical starting checklist

  1. Install Blender 3.0 or newer (the current release is 5.2.1 LTS) and check you have Python 3.10 or newer. From Blender 5.0 the macOS build is Apple Silicon only — on an Intel Mac install 4.5 LTS instead.

  2. Install the uv package runner with its official installer, never with pip: macOS `brew install uv`; Windows `powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"`; Linux `curl -LsSf https://astral.sh/uv/install.sh | sh`. On Windows, restart your AI app afterwards so it picks up the new PATH.

  3. Install the Blender add-on with one command: `uvx blender-mcp install-addon`. To do it by hand instead, download addon.py from the repository, then in Blender use Edit → Preferences → Add-ons → Install and enable "Interface: MCP for Blender".

  4. Point your AI app at Blender by adding a server named `blender` that runs the command `uvx` with the single argument `blender-mcp`. That goes in claude_desktop_config.json for Claude Desktop, or Settings → MCP in Cursor. On Claude Code you can skip the file: `claude mcp add --scope user blender -- uvx blender-mcp`, where the double dash is required.

  5. In VS Code only, the top-level key is `servers`, not `mcpServers`. Pasting the Claude Desktop block unchanged configures nothing at all and shows no error — open the file with the "MCP: Open User Configuration" command and use the right key.

  6. Open Blender, press N in the 3D viewport to show the sidebar, find the "MCP for Blender" tab, tick the integrations you want, and press the connect button (currently labelled "Connect to MCP server"). Then prompt from your AI app.

  7. Keep exactly one client connected. Everything talks over a single port, so do not run `uvx blender-mcp` yourself in a terminal and do not leave two AI apps connected at once — that is the most common reason it silently does nothing.

  8. Work in stages and save first. One instruction per prompt beats one large request, because each operation has a time limit; and the plugin executes AI-written Python inside Blender with your normal permissions, so start in a scratch .blend file rather than a project you care about.

Open the source, not a summary

The prompts

  1. 01 · Check the connection works

    Send this first, before anything creative.

    List every object currently in my Blender scene, then delete the default cube.
    
    If nothing happens, send it a second time — the first command after connecting often fails silently while the server is still starting.
  2. 02 · Block out the model

    One stage per prompt. Do not ask for the finished thing in one go.

    Build a low-poly kaiju monster: a big bipedal reptile with a thick heavy tail, jagged spikes down its back, short arms and thick legs. Block the whole thing out with simple shapes first, get the proportions right, and show me before you add any detail.
    
    Then follow up separately with "now refine the head and add the spikes", and again with "give it a rough grey-green scaly material".
  3. 03 · Environment and lighting

    Switch on Poly Haven in the add-on sidebar first.

    Use Poly Haven. Find an outdoor sunset HDRI and set it as the world background, then add a large ground plane and put a cracked concrete texture on it. Keep the monster where it is.
  4. 04 · Camera and a render you can post

    Ask for the render on its own — bundling heavy work is what hits the timeout.

    Add a camera at a low hero angle looking up at the monster so it fills about two-thirds of the frame, add a rim light behind it, set the render to 1920x1080, and render a preview image.