# Pantani/tdmcp [Health: Active]

**Category:** 🎨 Art & Culture  
**Repository:** https://github.com/Pantani/tdmcp  
**GitHub Stars:** 39  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/pantani-tdmcp

## Description
Stop wiring nodes by hand — describe a visual and the AI builds a real, playable TouchDesigner network: audio-reactive, generative, particle, 3D and feedback systems with live knobs and MIDI/OSC/DMX, checking and previewing its own work.

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "tdmcp": {
    "command": "npx",
    "args": ["-y","pantani-tdmcp"]
  }
}
```

## Documentation

## What Pantani/tdmcp MCP server does

Pantani/tdmcp MCP server connects an MCP-capable AI assistant to TouchDesigner. It translates requests such as creating a feedback tunnel or an audio-reactive particle system into actual TouchDesigner networks inside `/project1`. The generated networks can include nodes, connections, controls, and external I/O workflows for OSC, MIDI, DMX, and NDI.

The project includes reusable generators for feedback networks, audio-reactive systems, particle systems, and generative art, along with lower-level operations for creating, changing, inspecting, and arranging nodes. It can also create control panels, animate parameters, and return a thumbnail preview so an assistant can iterate on the result.

## How it works

Pantani/tdmcp MCP server has two local runtime parts. The MCP server is a Node program that exposes TouchDesigner tools and an embedded knowledge base to the AI client. The bridge runs inside TouchDesigner and performs the requested operations in the open project.

The workflow is designed around creation, verification, and preview. After the assistant asks for a visual, the server uses its reference material to select TouchDesigner operators and sends instructions through the bridge. The bridge creates and connects the nodes, checks the result for errors, and provides a preview. Generated networks are automatically arranged in a left-to-right layout.

The reference includes information on 629 operators, 68 Python classes, workflow patterns, GLSL techniques, and tutorials. This is useful when an assistant needs TouchDesigner-specific context rather than only generic code generation.

## Setup and configuration

Pantani/tdmcp MCP server requires TouchDesigner and an MCP-capable client. The README lists Claude Desktop, Claude Code, Codex, and Cursor as supported client options. TouchDesigner’s free non-commercial edition is sufficient. Node.js 20 or newer is needed when building from source; the Claude Desktop extension bundles the server and does not require a separate Node installation.

For Claude Desktop, install the released `.mcpb` extension and use the default bridge host and port: `127.0.0.1` and `9980`. For a source installation, clone the repository and run `npm run setup`, which installs and builds the project and prints the client connection line.

The bridge must also be installed in TouchDesigner. The documented options include dragging the release `.tox` package into `/project1` and clicking its install control, or running the provided bootstrap command from TouchDesigner’s Textport. The bridge starts on port `9980` by default.

Creative RAG is optional and disabled by default. It can be enabled with `TDMCP_RAG_ENABLED=1` and a local Ollama installation, then managed with the `tdmcp creative-rag` commands.

## Limitations and notes

The bridge executes arbitrary Python in the TouchDesigner process and listens on port `9980` on all interfaces by default. It should therefore be used only on a trusted network. For less trusted networks, the documentation recommends setting a bridge token with `TDMCP_BRIDGE_TOKEN` and/or disabling execution endpoints with `TDMCP_BRIDGE_ALLOW_EXEC=0`.

The AI client and TouchDesigner bridge are separate setup steps. After adding the MCP server, the AI client may need to be restarted. The bridge must be running before requests can create or inspect a network. Creative RAG is a local, opt-in feature and does not provide the bridge, DMX access, or Python execution itself.

_Full upstream README: https://allmcps.com/mcp/pantani-tdmcp/readme_