tdmcp
Specialized verticals · tested 2026-08-07 · re-test due 2026-11-07 · by the Hlido desk, not the vendor
In short: A TouchDesigner MCP server with a create→verify→preview loop and 629 real operators embedded — ambitious surface area for a solo project, and honest about which parts are experimental.
5 PASS · 0 FAIL of 5 public-surface claims
Quick answer
tdmcp scores 70/100 (STEADY) on Hlido’s independent, hands-on test (reviewed 2026-08-07). STEADY (70) for grounding the model in 629 real operators rather than letting it invent them, a create→verify→preview loop so the AI checks its own output, and genuinely performance-oriented features (MIDI/OSC/DMX, cues,
tdmcp lets you describe a visual — "a feedback tunnel from noise with blur and displace, then bloom, output to a window" — and have Claude, Cursor or Codex build a real, playable TouchDesigner network, wired and auto-arranged. The credibility rests on grounding rather than generation: an embedded reference of 629 operators, 68 Python classes, workflow patterns and GLSL techniques means the model uses operators that exist instead of inventing plausible-sounding ones, which is the standard failure mode when an LLM drives a specialist tool. A bridge inside TouchDesigner then creates, connects, inspects and previews nodes in a create→verify→preview loop and auto-arranges the network into a readable layout, so the AI checks its own work rather than asserting success. The live-performance orientation is concrete — control panels, presets, cues, tempo sync, MIDI/OSC/DMX I/O, a phone remote — and a local LLM copilot handles simple tasks without a paid API. 507 tools across three layers is a very large surface for what appears to be a solo project, and breadth at that ratio usually means uneven depth; the Creative RAG feature is labelled experimental, which is the right call and the kind of labelling we credit. Requires TouchDesigner, so the audience is narrow. No adoption figures or third-party validation. Reviewed from the public surface, not run against a real install.
Why STEADY
STEADY (70) for grounding the model in 629 real operators rather than letting it invent them, a create→verify→preview loop so the AI checks its own output, and genuinely performance-oriented features (MIDI/OSC/DMX, cues, tempo sync). Held below VITAL because 507 tools is a very large surface for an apparently solo project and implies uneven depth, there is no adoption evidence, and it requires TouchDesigner — narrow by construction.
Public-surface checklist
- PASS Homepage loads (required)
- PASS Primary value prop (required) — 'Stop wiring nodes by hand.'
- PASS Cta present (required) — 'I'm an artist — start here' / 'Developer reference'
- PASS Pricing or access — Open source; local copilot avoids paid API for simple tasks
- PASS Evidence or demo — Tools reference and guides; no third-party validation
What we saw
4 screenshots captured by the Hlido engine during the reviewed run (run-21e747018a75275d-pantani-github-io). Our own captures — not vendor marketing material.
What it does well
- Embedded reference of 629 operators, 68 Python classes, workflow patterns and GLSL techniques — no invented operators
- create → verify → preview loop with auto-arranged, readable network layout
- Built for live performance: control panels, presets, cues, tempo sync, MIDI/OSC/DMX I/O, phone remote
- Local LLM copilot handles simple tasks without a paid API
- Server stays usable when TouchDesigner is closed
- Labels the Creative RAG feature experimental rather than shipping it as finished
- Documentation offered in Portuguese as well as English
What it fails at
- 507 tools across three layers is very wide for an apparently solo project — depth will be uneven
- Requires TouchDesigner; audience is narrow by construction
- No adoption figures, named users, or third-party validation
- Not exercised against a real install — network correctness and visual quality are unassessed
- Creative output quality is subjective and cannot be scored from a public surface
Best for
- TouchDesigner artists who want to prototype networks conversationally
- Live visual performers needing playable, control-mapped systems generated quickly
- Developers exploring how to ground an LLM in a specialist tool's real object model
Not recommended for
- Anyone not using TouchDesigner
- Production show files where an unverified generated network would be risky
- Buyers needing evidence of adoption or support
Pricing & access
- Pricing findable on the public surfacePASS Open source; local copilot avoids paid API for simple tasks (tested 2026-08-07)
Derived from Hlido-held evidence only (engine checklist + editorial text); quotes are verbatim from the scorecard; not vendor-supplied; re-derived daily. Verify current prices on the vendor's pricing page. Last verified 2026-08-07.
Compared to
Agent relevance
CLI MCP Behavioral-testable
Agentic-Commerce Readiness 69/100 · INTEGRABLE
Independent readiness for agent delegation & transaction. How it’s scored · check live
MCP server driving a desktop application through a resident bridge — the agent is the builder. A local chat copilot covers simple tasks without an API key.
Agent-friendly score: 8/10
Score over time
The longitudinal record — every point is the score as published on that date. Raw series.
Evidence
- 507 tools across three layers for TouchDesigner network construction — source (2026-08-07) verified
- Embedded reference of 629 operators and 68 Python classes — source (2026-08-07) verified
- create → verify → preview loop with automatic network layout — source (2026-08-07) verified
- Live-performance I/O: MIDI, OSC, DMX, cues, tempo sync, phone remote — source (2026-08-07) verified
- Creative RAG explicitly labelled experimental and opt-in — source (2026-08-07) verified
- Adoption figures or third-party validation — source (2026-08-07)
scorecard.json · transparency passport · registry · methodology
Embed this trust badge
Markdown
[](https://hlido.eu/check/?agent=pantani-tdmcp)HTML
<a href="https://hlido.eu/check/?agent=pantani-tdmcp"><img src="https://hlido.eu/badge/pantani-tdmcp.svg" alt="Hlido trust score"></a>