Claude × Blender MCP: Building a Game Playthrough Video in (Less Than) One Morning

I’ve been exploring the intersection of LLMs and game development — specifically, how far we can push a natural-language-to-game pipeline without traditional game engines.

This morning I tried something fun: using Claude’s Blender MCP to build a demo for the goal playable concept Ownership. The result is a short playthrough video showing the game scene transitions and corresponding key bindings — all authored inside Blender, driven by Claude. I didn’t operate in Blender manually from the beginning to the end.

The catch

Blender is a modeling & rendering tool, not a game engine. So what you see above is essentially a scripted animation that simulates gameplay, not a real interactive build. Still, the speed is remarkable — concept to video in under a morning.

Where this is heading

I’m researching LLM-driven game engines and the broader NL → Game pipeline: can we go from a natural-language description of a game mechanic to a playable prototype, automatically?

If this sounds interesting, here are two papers where we dig deeper:

• Grounding Machine Creativity in Game Design Knowledge Representations: Empirical Probing of LLM-Based Executable Synthesis of Goal Playable Patterns under Structural Constraints
• Mage: Multi-Axis Evaluation of LLM-Generated Executable Game Scenes Beyond Compile-Pass Rate

More to come — stay tuned.