POPX with AI & MCP
Use POPX with AI assistants like Claude and ChatGPT: every operator carries a README written for language models.
POPX and AI Assistants
AI assistants such as Claude (Anthropic) and ChatGPT (OpenAI) can help you build, debug and explain TouchDesigner networks. Connected to TouchDesigner through MCP (Model Context Protocol), an assistant can read your network, change parameters and create operators for you.
An assistant only knows what it can see, and the inside of a POPX operator is a whole network of POPs, shaders and scripts. To give it the full picture, every POPX operator ships with a README: a text DAT inside the operator that explains what it does, in a form a language model can read in one go.
The README Inside Every Operator
Open any POPX operator (for example a Noise Falloff) and you will find a Text DAT named README next to its internal network. It is plain Markdown, so you can read it yourself too.
Every README follows the same structure, so an assistant knows where to look:
- → What it does: the operator’s purpose and behaviour in a few paragraphs.
- → Inputs and outputs: every input and output, what it expects and what it returns.
- → Parameters: what each parameter does and how parameters affect each other.
- → Attributes: the POP attributes the operator reads, writes and creates (for example popxFalloff, Transform, Color).
- → How it works: the internal network and the order things happen in.
- → Packed point (POPX Geometry): the shared data model POPX operators pass between each other, so the assistant understands how operators connect in a chain.
- → Notes: limits, edge cases and things to watch for.
- → Docs: a link to the operator’s page on this website.
Using It With MCP
With a TouchDesigner MCP server installed, your assistant can run Python inside TouchDesigner. Ask it to read the README of the operators you are working with before it makes changes. The README sits directly inside each operator, so one line of Python reads it:
op('noise_falloff1').op('README').textExample Requests
A good habit is to start a session with: “Read the README of every POPX operator in /project1 before you change anything.” The assistant then knows which attributes each operator expects, which parameters matter, and how the chain fits together. Some requests that work well:
- → “Read the README of rigid1 and explain why my bodies fall through the floor.”
- → “Build a chain that scatters points on a sphere, adds a Noise Falloff and drives a Transform Modifier with it. Read each operator’s README first.”
- → “Which attribute does this Field Falloff read? Check the README and wire my field to it.”
- → “Read the Aura README and set up a scene with a key light and a glass material.”
Without MCP
You can use the README with any chat assistant, no MCP needed. Open the README DAT inside the operator, copy its text, and paste it into your conversation with Claude, ChatGPT or any other assistant along with your question.
Pasting the READMEs of every operator in your chain gives the assistant the whole picture, and the answers are far more accurate than from general knowledge alone.
Tips
- → The README matches the version of the operator it lives in. After updating POPX, the README updates with the operator.
- → Stubbed operators have no internals, so they have no README either. Replace the stubs first (see Stubs & Updates).
- → Assistants can still make mistakes. When something looks off, check the operator’s page in the Operators documentation .
- → Ask the assistant to explain a change before it makes it. Short, focused requests work better than one large one.