Blender MCP setup with Claude, Codex and Gemini CLI

Learn to connect Blender MCP to three AI clients, test scene creation, and understand the speaker's local model limits on an RTX 5070 with 12 GB VRAM.

Player not loading? Watch on YouTube

This tutorial connects Blender's MCP server to Codex, Gemini CLI and Claude Desktop for natural-language scene editing. The walkthrough uses Windows and PowerShell; the metadata lists Blender 5.1, Python, Node.js, Git and UV as prerequisites, along with a supported AI client.

Setup begins with the Blender add-on. The speaker drags the installation button from the browser into Blender twice, then checks the add-ons list. Next comes cloning the server repository and running it with UV for the initial installation and build. If a missing docutils dependency causes an error, the suggested fix is to install it with UV pip and retry.

Each client needs a configuration entry pointing to the cloned server directory. The tutorial covers Codex's config.toml, Gemini's settings.json and Claude Desktop's configuration, then checks whether each client can identify objects in the active scene. Codex and Gemini create golden spheres. Claude turns a sphere into an Earth-like planet, though the speaker notes that its geography is inaccurate. Gemini also produces a medieval lakeside village made mostly of basic geometry.

The local LLM section reports context exhaustion after two or three messages on an RTX 5070 with 12 GB of VRAM. Increasing context still led to a crash after a sphere test, so the video omits a full local setup. The speaker reports a smoother experience with the hosted clients.