Blender MCP

ahujasid/blender-mcp connects Blender's Python API to Claude via the Model Context Protocol. Result: you describe a 3D scene in plain English, Claude generates and edits it inside a live Blender session, and you watch the geometry update in real time.This shipped as part of Anthropic's April 28, 2026 creative-tool MCP wave alongside Adobe Creative Cloud, Autodesk, Ableton, and Splice. It's the moment Claude went from text agent to creative-suite operator.What it actually doesThe server exposes the Blender Python API surface as MCP tools — add primitive, transform, parent, set material, render. Claude calls those tools the same way it would call any MCP server. Because Blender's API is Python, the server can also execute arbitrary scripts when you grant permission, which is how Claude does things like "arrange these 30 objects in a fibonacci spiral" without a hand-written tool for every operation.Installgit clone https://github.com/ahujasid/blender-mcp cd blender-mcp # Add to your Claude Desktop or CLI MCP config { "mcpServers": { "blender": { "command": "uv", "args": ["run", "blender-mcp"] } } }Then open Blender, run the bundled addon, and Claude can connect to your live session.Why this matters3D modeling has the steepest learning curve in the creative-software stack. A natural-language layer doesn't replace a TD, but it does let designers, marketers, and non-3D-native engineers prototype scenes in minutes instead of days. For ad creative, product visualization, and AR/VR previews, that's a real shift.Compare with Figma MCP in the 2D design domain, or the Dreamina consumer image stack — same direction (AI inside the creative tool, not next to it), different surface.For the broader creative-AI speed picture, see today's AI image generator speed ranking — the 2D side of the same creative-tool shift.LimitationsRequires Blender 4.0+ running locally; not a cloud serviceComplex animation rigging still needs human cleanupRenders happen on your hardware — heavy scenes are gated by GPU