@hybridlabor-api/aos 4.3.0 → 4.4.0

package/mcps/tdmcp/docs/guide/recipes.md

@@ -1,72 +0,0 @@
----
-description: "Recipe gallery for tdmcp, the TouchDesigner MCP server — ready-made visual systems you can build in TouchDesigner with a single prompt."
----
-
-# Recipe gallery
-
-Recipes are **pre-built, tested networks** you can ask for by name — a fast way to
-get a known-good starting point you can then tweak. Just say:
-
-> *"Apply the **performable feedback tunnel** recipe and show me a preview."*
-
-The AI builds it, checks it, and previews it. From there, iterate in plain
-language like any other visual.
-
-## Built-in recipes
-
-| Recipe | What it makes | Level |
-| --- | --- | --- |
-| **Feedback Tunnel** | A hypnotic tunnel that feeds a transformed, blurred frame back into itself over a noise seed. | Beginner |
-| **Performable Feedback Tunnel** | The same tunnel, but with live knobs for decay, zoom, spin and blur — ready to perform, animate with an LFO, or snapshot as presets. | Intermediate |
-| **Decks + Layer Mixer (VJ A/B)** | Two source decks (A/B) with per-deck gain knobs, crossfaded through a compositeTOP mixer — the VJ-deck pattern from `create_decks` + `create_layer_mixer`. | Beginner |
-| **Audio Spectrum Bars** | A classic spectrum analyzer: live audio drawn as colored frequency bars. | Beginner |
-| **Reaction Diffusion** | A Gray-Scott reaction-diffusion simulation running on the GPU — organic, growing patterns. | Advanced |
-| **Noise Landscape** | A 3D terrain displaced by noise, shaded and rendered with an orbiting camera. | Intermediate |
-| **Particle Galaxy** | A swirling galaxy of particles from a sphere emitter, pushed by turbulence and gravity, glowing sprites. | Advanced |
-| **Webcam Glitch** | Live webcam through edge detection, RGB split, a feedback loop and a glitch shader. | Intermediate |
-| **Atemporal Body-Track Glitch Timeline** | 9:16 source-video template with clean footage, short green bug cuts, red small-point/line tracking trails and a performable `SceneMode` switch. | Advanced |
-| **Data Sonification** | Reads a data table into CHOPs and maps the values to both audio and a color ramp. | Intermediate |
-| **Projection Mapping** | Corner-pin keystone correction and edge feathering, ready to send to a projector. | Advanced |
-| **LED Strip Mapper** | Samples a visual down to one pixel per LED and streams colors out to an addressable LED strip. | Advanced |
-| **Kinect Silhouette** | Uses a Kinect Azure depth feed to isolate a body silhouette and drive a glowing outline. *(Needs Kinect Azure hardware.)* | Advanced |
-| **Depth Displacement + Post** | Synthetic depth map warps a ramp source through a Displace TOP, then a post stack (blur + level grade) finishes it — runs with zero hardware. | Intermediate |
-| **Kinetic Text Path Follow** | Kinetic text gliding smoothly along a deterministic circular path (sin/cos LFOs animating tx/ty on the Transform TOP) — placeholder until the native path-follow extension lands. | Intermediate |
-| **Kinetic Text Audio Reactive** | Manual-wiring template: kinetic text whose brightness pulses with the live audio bass band. Requires a one-line expression binding after import. | Intermediate |
-| **Optical Flow Particles** | CPU optical-flow (frame-diff + blur) drives a flow texture; the artist wires it into the particle render path after import. | Advanced |
-| **MediaPipe Face Overlay** | Manual-wire template: applies MediaPipe face landmarks (from `setup_face_tracking`) as a dot-cloud overlay over a live video source. | Intermediate |
-| **Scene Timeline Demo** | Declarative timeline with 3 scenes and crossfade, showcasing the `create_scene_timeline` Layer-1 orchestrator. | Intermediate |
-| **3D Scene Basic** | Foundational `create_3d_scene` template: geometry + camera + light + render pipeline — the starting point for 3D visuals. | Intermediate |
-| **Video Synth Oscillator** | Procedural Lissajous oscillator color synth from `create_video_synth` — analog video-synth look with live Speed / Freq / Color controls. | Beginner |
-| **Kinetic Text (Standalone)** | Text-only showcase of `create_kinetic_text` styles (Transform + Level + LFO) without any audio binding — pure typography pulse. | Beginner |
-| **Feedback Network Basic** | Minimal feedback TOP + level + composite pattern, the classic recursive look from `create_feedback_network`. | Beginner |
-| **GLSL Shader Basic** | Single GLSL TOP with inline shader showcasing `create_glsl_shader` with exposed uniform controls. | Intermediate |
-| **Pose Skeleton (Standalone)** | Placeholder skeleton renderer with instanced joints/bones, foundation for any custom pose source (no MediaPipe). | Intermediate |
-| **Particle System Basic** | Foundational `create_particle_system` template with grid emitter + force CHOP + geometry render. | Beginner |
-| **Audio Reactive Basic** | Minimal audio-in → analyze pattern from `create_audio_reactive`, ready for the artist to wire into a visual chain. | Beginner |
-| **Keyframe Animation Basic** | Animation COMP showcase with 2 channels + 5 keyframes, foundation for declarative camera/object motion. | Intermediate |
-
-::: tip Browse them live
-Ask *"list the available recipes"* and the AI will read them straight from the
-installed version, including any custom recipes you've saved to your
-[Obsidian vault](/reference/tools#obsidian-vault).
-:::
-
-## Hardware recipes
-
-A few recipes need extra gear:
-
-- **Kinect Silhouette** needs a **Kinect Azure** sensor.
-- **LED Strip Mapper** needs an **addressable LED strip** on a serial connection.
-- **Projection Mapping** assumes you have a **projector** to send the window to.
-
-They'll still build without the hardware so you can see the structure, but the
-live input/output won't do anything until the device is connected.
-
-## Save your own
-
-Made a look you love? Ask:
-
-> *"Save this network as a reusable recipe."*
-
-With an [Obsidian vault](/reference/tools#obsidian-vault) configured, it's stored
-as a Markdown note and shows up next to the built-ins for next time.

package/mcps/tdmcp/docs/guide/session-profile.md

@@ -1,90 +0,0 @@
----
-description: "Teach tdmcp your taste in TouchDesigner — a persistent session profile plus corpus-learning tools that distil your palettes, naming and favourite generators so the AI builds in your style."
----
-
-# Session profile & corpus learning
-
-By default every conversation starts cold. This arc lets tdmcp **remember your
-taste** across sessions: a persistent profile the AI loads in one read, fed by
-tools that learn your conventions from a live project, from your own library of
-notes, or from a standing record of your preferences.
-
-Reach for this when you find yourself repeating the same instructions — "always
-magenta on black", "name containers in snake_case", "I prefer the particle flock
-over the GPU field" — and want the assistant to start from your style instead of
-a generic default.
-
-## The session profile
-
-**`load_session_profile`** (the AI-layer tool) reads or initialises
-`~/.tdmcp/session-profile.json` — a cross-session snapshot that caches the latest
-outputs of the learning tools below. It returns a unified JSON with a
-`loaded_at` timestamp and `style_memory`, `recent_work`, `conventions` and
-`corpus_style` sections, creating sensible defaults if no file exists. Override
-the path with `TDMCP_SESSION_PROFILE_PATH`.
-
-The same data is exposed read-only as the **`tdmcp://session/profile`** MCP
-resource, so an agent can pull all your preferences in a single resource read at
-the start of a turn.
-
-> *"Load my session profile so you know my style before we build."*
-
-## The learning tools
-
-These four tools (the vault group — they need an
-[Obsidian vault](/reference/tools#obsidian-vault) via `TDMCP_VAULT_PATH`)
-populate the profile. None of them mutate your TouchDesigner project:
-
-- **`style_memory`** reads, updates or summarises a standing `Memory/style.md`
-  note — your long-lived record of palettes, default energy, banned moves,
-  favourite generators and naming/layout conventions. Modes: `show` (a compact
-  one-line context for the LLM), `read` (full structured), `update` (field-wise
-  merge).
-- **`learn_conventions`** walks a live TouchDesigner subtree read-only and
-  extracts your house conventions — naming case, colour tags, container shapes,
-  parameter defaults, layout direction — writing them to `Memory/conventions.md`
-  and optionally merging confident signals into `Memory/style.md`. No TD
-  mutations.
-- **`learn_from_my_corpus`** is the offline companion: it walks your vault corpus
-  (Recipes, Components, Looks, Setlists, Moodboards) and distils palette, naming,
-  recipe-shape and parameter preferences into `Memory/corpus_style.md`. No
-  TouchDesigner required — pure filesystem read.
-- **`recall_similar_work`** ranks your memory notes by similarity to a visual
-  goal (query tokens plus tag/operator overlap) and returns the closest prior
-  recipes, params and prompts so the agent can reuse them.
-
-> *"Learn my conventions from `/project1`, then learn from my vault corpus, and
-> update my style memory with anything you're confident about."*
-
-## A typical flow
-
-1. Configure a vault (`TDMCP_VAULT_PATH`) and build a few things you like.
-2. Run `learn_conventions` on a project you're proud of and `learn_from_my_corpus`
-   over your saved looks; both write to `Memory/*.md`.
-3. Run `load_session_profile` to fold those into `~/.tdmcp/session-profile.json`.
-4. In future sessions, the AI reads `tdmcp://session/profile` (or you say *"load
-   my profile"*) and builds in your style from the first prompt.
-
-## Related: the optional RAG libraries
-
-Corpus learning above is about *your own* taste. tdmcp also ships two opt-in,
-local-only retrieval libraries that broaden the well the AI draws from — both
-gated behind `TDMCP_RAG_ENABLED=1` and off by default:
-
-- **Creative RAG** — a curated repertoire of techniques and references, exposed
-  as `tdmcp://creative/cards/{id}` and `tdmcp://creative/search`. See
-  [Creative RAG](/creative-rag).
-- **Project RAG** — a local index of TouchDesigner projects/components (also
-  needs `TDMCP_PROJECT_RAG_ENABLED=1`), exposed as `tdmcp://project/cards/{id}`,
-  `tdmcp://project/search` and `tdmcp://project/sources`. See
-  [Project RAG](/project-rag).
-
-Both are read-only and always carry source URL, license and rights notes on every
-result. They are documented in full on their own pages; this guide covers the
-always-on session-profile path.
-
-## See also
-
-- [Working from your own notes (Obsidian vault)](/guide/prompt-cookbook#working-from-your-own-notes-obsidian-vault)
-  in the prompt cookbook.
-- [MCP resources](/guide/mcp-resources) for the full `tdmcp://…` resource map.

package/mcps/tdmcp/docs/guide/shader-park.md

@@ -1,94 +0,0 @@
----
-description: "Use Shader Park with tdmcp: compile Shader Park sculpture code into a TouchDesigner GLSL MAT network, download the official Shader Park .tox plugin, and build an Animus-inspired native audio visualizer."
----
-
-# Shader Park and visualizer packs
-
-tdmcp supports two practical paths for Shader Park:
-
-| Path | Best for | How |
-|------|----------|-----|
-| tdmcp tool | AI-built scenes, repeatable scripts, previews | `create_shader_park` / `tdmcp-agent shaderpark` compiles code with `shader-park-core` and builds a GLSL MAT render network. |
-| Official plugin | Hands-on Shader Park editing inside TouchDesigner | `npm run shader-park:tox` downloads the official `Shader_Park_TD.tox` from `shader-park-touchdesigner`. |
-
-The `codygibb/animus-visualizer` repo is different: it is a Processing/Java
-sketch, not an npm package or TouchDesigner component. tdmcp includes a native
-recipe inspired by its audio-ring style, so you can build the idea without
-installing Processing.
-
-## Compile Shader Park code
-
-`create_shader_park` uses the optional `shader-park-core` peer dependency. The
-default hosted/package install does not include it because `shader-park-core@0.2.8`
-publishes a broken `toMinimal` bin entry that makes pnpm emit install warnings.
-For local Shader Park compilation, install it next to tdmcp first:
-
-```bash
-npm install shader-park-core
-```
-
-Ask your assistant:
-
-> *"Create a Shader Park sculpture with `rotateY(time * 0.25); color(vec3(0.2, 0.8, 1.0)); sphere(0.45);`, expose the controls, and show me a preview."*
-
-Or run it from the shell:
-
-```bash
-tdmcp-agent shaderpark --params '{
-  "code": "let size = input(); rotateY(time * 0.2); sphere(size);",
-  "uniform_values": { "size": 0.55 },
-  "speed": 1,
-  "scale": 1,
-  "camera_z": 4
-}'
-```
-
-The tool creates a new COMP with:
-
-- a `shaderpark_code` Text DAT holding the original Shader Park source
-- a compiled `shaderpark_pixel` Text DAT
-- a `glslMAT` assigned to a renderable bounding box
-- camera, light, Render TOP and `out1`
-- live controls for Speed, Scale, Opacity, StepSize, CameraZ and any float
-  `input()` uniforms, such as `Size`
-
-Use `uniform_values` when your Shader Park code contains `input()` values. For
-example, `let size = input(); sphere(size);` needs `{"size": 0.55}` or the
-uniform starts at Shader Park's default value.
-
-## Download the official `.tox`
-
-Shader Park also ships a ready-made TouchDesigner plugin. Fetch it with:
-
-```bash
-npm run shader-park:tox
-```
-
-That downloads `Shader_Park_TD.tox` into:
-
-```text
-vendor/shader-park/Shader_Park_TD.tox
-```
-
-The file is intentionally ignored by git. Drop it into TouchDesigner when you
-want the official plugin workflow: a Text DAT with Shader Park code connected to
-the plugin. If your code uses `input()`, add a matching uniform inside the
-plugin's GLSL MAT, following the upstream instructions.
-
-## Animus-style rings
-
-Build the native recipe:
-
-```bash
-tdmcp-agent recipe --params '{"id":"animus_rings_visualizer"}'
-```
-
-It uses a synthetic `audiooscillatorCHOP`, converts the channel to a TOP, and
-draws pulsing radial rings in a GLSL TOP. Swap the oscillator for Audio Device In
-or Audio File In when you want live music.
-
-## Sources
-
-- [shader-park-core](https://github.com/shader-park/shader-park-core)
-- [shader-park-touchdesigner](https://github.com/shader-park/shader-park-touchdesigner)
-- [animus-visualizer](https://github.com/codygibb/animus-visualizer)

package/mcps/tdmcp/docs/guide/show-timelines.md

@@ -1,125 +0,0 @@
----
-description: "Compose and automate a live show in TouchDesigner with tdmcp — timelines, setlists, beat-locked cue sequencing, and a master safety blackout, all from plain language."
----
-
-# Show timelines & setlists
-
-A single look is a moment; a *show* is the whole night. This arc is the
-compose-and-automate layer: scaffold a show container, store and recall cues,
-roll through a setlist on the wall clock, sequence cues to the beat, and keep a
-master safety blackout one button away. You describe the order and the feeling;
-tdmcp builds the timers, transitions and controls.
-
-Reach for these tools when you're moving from *making visuals* to *running a
-set* — a DJ night, a VJ slot, a theatre cue stack, an installation that has to
-loop unattended.
-
-## Start with a stage
-
-`scaffold_show` creates the blank skeleton: a new container with a `master`
-output Null (where your mix lands) and a `tempo` beat clock for reactivity — but
-no scenes yet. It's the canvas you hang everything else on:
-
-> *"Scaffold a show called `mainstage` and show me the container."*
-
-From there you add looks, store them as cues, and wire the transport.
-
-## Cues: store, recall, morph
-
-`manage_cue` is the foundation of every cued workflow. It stores, recalls,
-morphs, lists and deletes **cues** — snapshots of a COMP's custom parameters,
-kept in the COMP's own storage. A morph can be eased and quantized to the beat
-or bar:
-
-> *"Store the current look as cue `drop`, then morph to it over 2 bars when I
-> recall it."*
-
-Almost everything below recalls from `manage_cue`: the sequencer, the timeline,
-the clip launcher and the set navigator all read the same cue store.
-
-## Timelines
-
-- **`create_scene_timeline`** (Layer 2) builds a scrubbable, timer-driven
-  playhead through an ordered list of scenes. Each scene points at a stored cue
-  and morphs into the next on its boundary, in seconds or bars. The
-  **Scene Timeline Demo** recipe shows it with three scenes and a crossfade.
-- **`control_timeline_transport`** (Layer 3) is the atomic transport verb:
-  play, pause, seek to a frame, jump to a named cue, or set the playback rate —
-  the hook an operator or the local copilot uses to drive playback.
-
-> *"Build a 3-scene timeline that crossfades intro → build → drop over 8 bars
-> each, then play it."*
-
-## Setlists
-
-- **`create_setlist_runner`** (Layer 1) is a wall-clock setlist player: rows of
-  (source TOP, duration, transition) auto-advance on a Timer CHOP with crossfade
-  or hard cut, an optional HUD overlay, and live Play / Row / Skip / Prev / Loop
-  controls. This is the "play these clips in this order, this long each" tool.
-- **`compose_cue_list`** (Layer 1) turns a natural-language show description into
-  a validated cue list. It uses the local LLM when configured, or a deterministic
-  grammar parser otherwise, and can chain straight into the sequencer with
-  `apply=true`.
-
-With an [Obsidian vault](/reference/tools#obsidian-vault) configured, you can
-also persist setlists as notes: **`import_setlist`** loads a setlist note and
-builds each track's recipe into TD, and **`export_setlist_to_vault`** writes the
-COMP's stored cues back out as a setlist note.
-
-> *"Run a setlist: clip A for 30 s, clip B for 45 s, clip C for 60 s, crossfade
-> 2 s between each, loop at the end, and show the HUD."*
-
-## Beat-locked sequencing & cue composition
-
-When the order should follow the music rather than the clock:
-
-- **`create_cue_sequencer`** (Layer 2) plays ordered steps (each a cue plus a
-  bar/beat count) quantized to the global tempo, with live Step / Active / Rate /
-  Loop controls.
-- **`create_phrase_locked_cue_engine`** (Layer 1) queues incoming cue pulses FIFO
-  and fires them on the next phrase boundary (1/2/4/8/…/64 bars), so a hit
-  triggered anywhere lands cleanly on the next phrase.
-- **`create_set_navigator`** (Layer 1) is a hands-light, QLab-style stage
-  navigator — step through an ordered list of scene/cue names with Index / Next /
-  Prev / Go controls, recalling each via `manage_cue`.
-- **`create_clip_launcher`** (Layer 2) lays cues out as a rows × cols button grid
-  that recalls or morphs on tap.
-- **`create_scheduler`** (Layer 2) fires named timers with segments — recall a
-  cue, set a parameter, or run a script stub — at each segment boundary.
-
-To lock all of this to an external transport, **`sync_external_clock`** (Layer 1)
-wires MIDI, OSC, Ableton Link or a network tempo into a Beat CHOP, so the
-sequencer, phrase-lock engine and quantized morphs all follow the incoming clock.
-
-## Live safety
-
-A show needs a kill switch that always works:
-
-- **`create_safety_blackout_chain`** (Layer 1) protects the master output with a
-  deterministic fade-to-black (configurable curve and time), an optional
-  hard-cut emergency snap, an optional hotkey and external watchdog trigger, and
-  a symmetric fade-in recovery. It is fully parameter-driven — no Python at cook
-  time — so it stays safe even with `TDMCP_BRIDGE_ALLOW_EXEC=0`.
-- **`create_panic`** (Layer 2) adds a per-source kill + freeze: a Blackout toggle
-  that drives brightness to zero and a Freeze toggle that holds the last frame.
-
-> *"Add a safety blackout on the master output with a 1.5 s ease-out fade and an
-> armed emergency hard-cut on a hotkey."*
-
-## How it fits together
-
-`scaffold_show` gives you the master output and tempo clock. `manage_cue` stores
-the looks. A **timeline** (`create_scene_timeline`) or **setlist**
-(`create_setlist_runner`) drives the order; a **sequencer**
-(`create_cue_sequencer` / `create_phrase_locked_cue_engine`) locks it to the
-beat via `sync_external_clock`; and `create_safety_blackout_chain` sits last on
-the master so the kill switch is always in reach.
-
-## See also
-
-- [Live performance & control](/guide/prompt-cookbook#live-performance-control)
-  in the prompt cookbook for copy-paste prompts.
-- [Front-of-house dashboard](/guide/dashboard-foh) to drive all of this from a
-  phone or touchscreen.
-- [Recipe gallery](/guide/recipes) for the Scene Timeline Demo and other
-  ready-made starters.

package/mcps/tdmcp/docs/guide/troubleshooting.md

@@ -1,121 +0,0 @@
----
-description: "Fix common tdmcp setup issues — connect the TouchDesigner MCP server and the bridge, and get your AI client talking to TouchDesigner."
----
-
-# Troubleshooting
-
-Most problems are one of a handful of things. Find what you're seeing below.
-
-## "TouchDesigner isn't reachable"
-
-The AI can't find the bridge. Check, in order:
-
-1. **Is TouchDesigner open?** It needs to be running.
-2. **Did you turn on the bridge?** In TouchDesigner's Textport you should have seen
-   `[tdmcp] bridge running on port 9980`. If not, redo
-   [step 3 of the install](/guide/install#turn-on-the-bridge).
-3. **Quick test:** open a terminal and run
-   `curl http://127.0.0.1:9980/api/info` — it should return some JSON. If it does,
-   the bridge is fine and the issue is on the AI-client side (next item).
-
-## The AI doesn't list any tdmcp tools
-
-The client didn't load the server.
-
-- **Restart your AI client** after installing/enabling the extension. This is the
-  most common fix — the tools only appear after a restart.
-- On Claude Desktop, confirm the "TouchDesigner (tdmcp)" extension is **enabled**
-  in **Settings → Extensions**.
-
-## The local copilot says "LLM offline"
-
-`tdmcp chat` talks to a local LLM (Ollama by default). If the header shows
-**LLM offline** or messages fail with *fetch failed*:
-
-1. **Restart `tdmcp chat`** (Ctrl-C, run it again). It **auto-starts Ollama** for
-   you — so a fresh launch brings the daemon back up if it had stopped.
-2. **Is Ollama installed?** It must be on your PATH — get it at
-   [ollama.com](https://ollama.com). (Homebrew's Ollama isn't a persistent daemon
-   and can stop between sessions; `tdmcp chat` starting it for you covers that.)
-3. **Model not pulled?** If the status says *model not pulled*, click **Pull** in
-   the UI or run `ollama pull qwen2.5:3b`.
-4. **Started with `--no-ollama`?** Then start the daemon yourself: `ollama serve`.
-5. **Using a remote / cloud endpoint?** Check `TDMCP_LLM_BASE_URL` and
-   `TDMCP_LLM_API_KEY` in your [environment](/reference/environment).
-
-## macOS microphone & camera permission {#macos-microphone-camera-permission}
-
-The **first time** a visual uses your microphone (audio-reactive) or camera
-(webcam), macOS shows a permission dialog.
-
-- **Click Allow.** Until you respond, TouchDesigner can look **frozen** (it's
-  waiting on the popup, sometimes with high CPU). That's expected — it isn't a
-  crash.
-- If you accidentally clicked **Deny**, fix it in
-  **System Settings → Privacy & Security → Microphone** (or **Camera**), enable
-  TouchDesigner, and restart it.
-- **Don't want the popup while testing?** Ask for a **test tone** source instead of
-  the microphone: *"use a test oscillator instead of the mic."*
-
-## The download / Textport line shows a network error
-
-Both the `.mcpb` download and the one-line bridge installer need to reach the
-internet (GitHub).
-
-- **Reconnect to the internet** and try again.
-- **Download link 404s?** A release may not be published yet — ask whoever shared
-  tdmcp for the `tdmcp.mcpb` file directly and
-  [install from the file](/guide/install#install-from-file).
-
-## "Port 9980 is already taken"
-
-Something else is using that port. Use a different one in **both** places:
-
-- In TouchDesigner: `from mcp import install; install.run(port=9981)`
-- In your AI client's settings: set the TouchDesigner **port** to `9981` (or the
-  `TDMCP_TD_PORT` [environment variable](/reference/environment)).
-
-## A visual built, but looks off
-
-The audio / particle / 3D builders and the more exotic recipes use best-effort
-parameter names and may need a nudge. Just say what's wrong:
-
-- *"The particles aren't moving — check it for errors and fix them."*
-- *"Explain what this network does so I can see what's missing."*
-
-## It runs slowly
-
-> *"This is running slow — find the bottleneck and lower the resolution where it
-> won't hurt."*
-
-The AI can measure cook times and optimize the heaviest parts.
-
-## TDAbleton Mapper is not moving Ableton
-
-For the MediaPipe hands to Auto Filter flow, AbletonMCP is not part of the runtime
-path. The path is TouchDesigner MediaPipe hands -> TDAbleton `TDA_Mapper` ->
-Ableton mapped parameter.
-
-Check, in order:
-
-1. TouchDesigner is playing; MediaPipe capture does not update while the timeline
-   is paused.
-2. The hand adapter CHOP has non-zero `confidence` and `handedness`.
-3. `/project1/hand_ableton_mapper/mapper_send` has moving `map1`, `map2`, `map3`,
-   and `map4` channels.
-4. Left hand moves `map1`/`map3`; right hand moves `map2`/`map4`.
-5. The active `TDA_Mapper` path is the real track/device target, not a stale
-   mapper from another track.
-6. `Oscinputchop` points at `mapper_send`, `Reorder` is `map1 map2 map3 map4`,
-   `Bypass1..4` are off, and `Min/Max1..4` are `0..1`.
-7. The four TDAbleton mapper slots are manually mapped to Auto Filter or rack macro
-   parameters in Ableton.
-
-Run `diagnose_tdableton_mapper` to inspect the mapper state from TouchDesigner and
-use `repair:true` only when you want tdmcp to restore the input CHOP, reorder,
-bypass, and range parameters.
-
-## Still stuck?
-
-Open an issue at [github.com/Pantani/tdmcp/issues](https://github.com/Pantani/tdmcp/issues).
-Developers: the [reference docs](/reference/architecture) cover deeper diagnostics.

package/mcps/tdmcp/docs/guide/what-is-tdmcp.md

@@ -1,62 +0,0 @@
----
-description: "tdmcp is the TouchDesigner MCP server — an AI assistant builds real TouchDesigner networks from your plain-language description, then checks and previews them."
----
-
-# What is tdmcp?
-
-**tdmcp lets you build visuals in [TouchDesigner](https://derivative.ca) just by
-describing them to an AI assistant.** You type what you want in plain language;
-the AI builds the actual network of nodes inside your project, checks it for
-errors, and shows you a preview.
-
-> *"Create a feedback tunnel from noise with blur and displace, then add bloom and
-> output it to a window."*
-
-…and the nodes appear, wired up, in your project — ready to tweak and perform.
-
-You don't need to know how to code, and you don't need to know which TouchDesigner
-operators to use. That's the AI's job. You stay in the role you care about:
-**directing the look and the feel.**
-
-## Who it's for
-
-- **VJs and live performers** who want to spin up audio-reactive, generative or
-  particle systems fast — already playable, with knobs to tweak live.
-- **Visual artists and installation makers** who'd rather describe an idea than
-  wire a hundred nodes by hand.
-- **Anyone learning TouchDesigner** who wants a working, correct network to learn
-  from and pull apart.
-
-If you *are* a developer and want the internals, head to the
-[developer reference](/reference/architecture).
-
-## Why it works
-
-Most "AI builds your project" tools guess. tdmcp doesn't, because it pairs two
-things:
-
-- **Real knowledge.** It carries a built-in reference of TouchDesigner's actual
-  operators, so the AI uses real nodes instead of inventing ones that don't exist.
-- **Real execution.** A small **bridge** runs inside TouchDesigner and actually
-  creates, connects and previews nodes — so the AI can *see* its own work, catch
-  mistakes, and fix them before handing it back to you. Every network it builds is
-  tidied into a clean left-to-right layout instead of a tangle.
-
-## What you'll need
-
-- **[TouchDesigner](https://derivative.ca/download)** — the free non-commercial
-  edition is fine.
-- An AI assistant that supports it: **Claude Desktop** (easiest — no terminal),
-  Claude Code, Codex, or Cursor. No paid API? tdmcp also ships a free
-  [local copilot](/guide/local-copilot) that runs a model on your own machine.
-
-## Next steps
-
-1. [Install it (Claude Desktop)](/guide/install) — about 3 minutes, no terminal.
-2. [Make your first visual](/guide/first-visual).
-3. Keep a [prompt cookbook](/guide/prompt-cookbook) and the
-   [recipe gallery](/guide/recipes) handy for ideas.
-
-::: tip Portuguese
-This documentation is also available in Portuguese — [start here](/pt/).
-:::

package/mcps/tdmcp/docs/index.md

@@ -1,49 +0,0 @@
----
-layout: home
-title: tdmcp — TouchDesigner MCP server
-titleTemplate: false
-description: "The TouchDesigner MCP server. Connect Claude, Cursor, or Codex to TouchDesigner and build real visual systems from plain language — it uses real operators and previews its own work."
-
-hero:
-  name: tdmcp
-  text: The TouchDesigner MCP server
-  tagline: Stop wiring nodes by hand. Describe a visual to Claude, Cursor or Codex and it builds a real, playable network inside TouchDesigner — audio-reactive, generative, particle and 3D systems with live controls — checking and previewing its own work.
-  actions:
-    - theme: brand
-      text: I'm an artist — start here
-      link: /guide/what-is-tdmcp
-    - theme: alt
-      text: Developer reference
-      link: /reference/architecture
-    - theme: alt
-      text: 🇧🇷 Read in Portuguese
-      link: /pt/
-
-features:
-  - title: Describe it, don't wire it
-    details: "\"Create a feedback tunnel from noise with blur and displace, then add bloom and output it to a window\" — and the nodes appear, wired up, in your project."
-  - title: Real operators, not guesses
-    details: An embedded reference of 629 operators, 68 Python classes, workflow patterns and GLSL techniques means the AI uses real TouchDesigner operators instead of inventing them.
-  - title: Sees and fixes its own work
-    details: A bridge inside TouchDesigner creates, connects, inspects and previews nodes in a create → verify → preview loop, and auto-arranges every network into a readable layout.
-  - title: Built for live performance
-    details: Audio-reactive, generative and particle systems arrive playable — with control panels, presets, cues, tempo sync, MIDI/OSC/DMX I/O and a phone remote.
-  - title: Works offline-friendly
-    details: A local LLM copilot (tdmcp chat) handles simple tasks without a paid API, and the server stays usable even when TouchDesigner is closed.
-  - title: 332 tools, three layers
-    details: From one-line artist generators down to atomic node CRUD, library packaging, AI session memory, vault workflows and Python escape hatches — see the full, always-current Tools reference.
-  - title: Creative RAG (experimental)
-    details: "Opt-in local creative repertoire — open-licensed artworks, artists and techniques you can search for inspiration. Off by default. Repertoire, not policy; no bridge, DMX or Python exec."
-    link: /CREATIVE_RAG
-    linkText: Read the Creative RAG guide
----
-
-## Two ways in
-
-**🎛️ I make visuals.** You don't need to code. Start with [What is tdmcp?](/guide/what-is-tdmcp),
-do the [one-click install](/guide/install), then [make your first visual](/guide/first-visual).
-Prefer Portuguese? [Comece por aqui »](/pt/)
-
-**🛠️ I'm a developer.** Jump to the [architecture overview](/reference/architecture), the
-auto-generated [tools reference](/reference/tools), or wire it into your client and read the
-[bridge & REST API](/reference/bridge-api).