One of the more interesting creative-coding tools to surface this year is Embody, created by Dylan Roscover: a full MCP (Model Context Protocol) implementation for TouchDesigner that lets an AI agent like Claude Code build TouchDesigner networks through conversation alone. Instead of manually dragging and wiring operators, you describe what you want and the agent assembles the node graph for you. Just as significant, Embody introduces TDN (TouchDesigner Network) — an open, JSON-like format for representing networks that is flexible, human-readable, and, crucially, diffable. And it’s completely free.
Why the file format is the sleeper feature
TouchDesigner’s power has always come with a workflow tax: its projects are binary-ish .toe files that don’t play nicely with Git, so meaningful version control, diffing, and collaboration have been awkward for years. A JSON-like network format changes that. If you can represent a network as readable text, you can diff two versions, review changes, merge collaborators’ work, and track a project’s history the way software developers take for granted. For teams and studios building large interactive installations in TouchDesigner, that’s arguably a bigger deal than the AI hook — it brings a mature engineering discipline to a tool that has resisted it.
What the AI layer actually enables
The MCP integration is the flashier half. By exposing TouchDesigner’s network to an AI agent, Embody lets you scaffold a patch conversationally — “make a feedback loop that reacts to audio,” “add a particle system driven by this CHOP” — and have the operators created and connected for you. That won’t replace the deep, hands-on knowledge that separates a TouchDesigner artist from a novice; node-based visual programming is a craft, and knowing what to ask for is most of the skill. But as a scaffolding and learning aid, having an agent that can build a working starting point is a real accelerant, especially for people climbing the notoriously steep early learning curve.
The bigger pattern
Embody is a concrete example of a trend this site keeps returning to: AI agents reaching into creative software through open protocols like MCP, acting as collaborators inside the tools artists already use rather than as separate generators that spit out finished files. It pairs naturally with TouchDesigner’s growing community-tools ecosystem — machine-learning tracking plugins, real-time 2D lighting operators — that keeps extending the platform. For creative coders, the takeaway is twofold: a long-wished-for diffable format has arrived, and conversational network-building is now something you can actually try, for free.