tdmcp: An MCP Server for Natural Language-Driven Visual Creation in TouchDesigner

tdmcp is a Model Context Protocol (MCP) server designed for TouchDesigner. It allows developers to describe visual needs in natural language, and AI automatically builds runnable node networks in TouchDesigner—supporting effects like audio response, generative art, particle systems, and 3D scenes. This tool transforms the workflow by letting users focus on creative intent while AI handles technical implementation.

Previous AI tools for visual creation lacked two critical elements:

1. True Knowledge Base: tdmcp includes a complete reference of 629 TouchDesigner operators, 68 Python classes, workflows, GLSL tips, and tutorials—so AI uses real, accurate components instead of guesswork.
2. Real Execution Ability: Through a TouchDesigner bridge component, AI can create, connect, check, and preview nodes. It uses a "create→verify→preview" loop to fix errors, and auto-arranges networks left-to-right for readability.

The system has three parts working together:

• AI Assistant: Where users input natural language (supports Claude Desktop, Claude Code, Codex, Cursor).
• tdmcp Server: A Node.js program providing TouchDesigner's toolset and operator knowledge (install once).
• Bridge Component: Runs inside TouchDesigner to let the server drive it (enable once per machine). Flow: User → AI → tdmcp Server → TouchDesigner (creates nodes in /project1).

Prerequisites: TouchDesigner (free non-commercial ok), MCP-supported AI (Claude Desktop easiest), Node.js 20+ (only for source build). Installation:

• Claude Desktop: Drag tdmcp.dxt into Extensions settings (keep host/port as 127.0.0.1:9980).
• Source Build: Clone repo → npm run setup. Start Bridge: In TouchDesigner Textport, run: import urllib.request; exec(urllib.request.urlopen("https://raw.githubusercontent.com/Pantani/tdmcp/main/td/bootstrap.py").read().decode()) (check for success message).

tdmcp offers over 102 tools across three layers:

• Artist-Friendly: One-click generation (feedback networks, audio reactive effects, particle systems, generative art).
• Building Blocks: Control panels, parameter animation, external I/O (OSC/MIDI/DMX/NDI).
• Atomic: Node CRUD, checks, property reading. Additional features: Real-time parameter adjustment, preset save/load, MIDI mapping, auto error checking, and preview thumbnails.

Example workflow:

1. Input: "Create an audio-responsive particle galaxy and preview it."
2. AI steps: Analyze需求 → select operators → create nodes → connect → check errors → return preview.
3. Iterate: Adjust with commands like "Make it warmer" or "Add feedback trails".

The bridge component runs Python code in TouchDesigner and listens on port 9980. To stay safe:

• Run only on trusted networks.
• For untrusted networks: Enable bridge authentication (use TDMCP_BRIDGE_TOKEN) or disable execution endpoints (TDMCP_BRIDGE_ALLOW_EXEC=0). See official docs for detailed security guidelines.

tdmcp represents a new AI-assisted creative paradigm:

• Deep Domain Integration: Structured knowledge of 629 operators for accurate AI output.
• Closed-Loop Feedback: AI can verify and preview its work.
• User Control: Users retain creative intent while AI handles tech.
• Progressive Complexity: Suits both beginners (one-click) and experts (fine control). For TouchDesigner users: Lowered entry barrier. For devs: A model of MCP integration with professional software.