Luxaudica speaks MCP — the open standard that lets AI assistants like Claude use real tools. Connect once, and the assistant becomes a VJ who can see the screen, hear the state of the desk, move every control, and build entirely new scenes — while you talk to it like a colleague. This page is the complete guide to that surface.
An assistant connected to Luxaudica isn't pasting keystrokes — it has the same first-class controls you do, plus judgment tools you don't: it can privately preview any look at four musical moments before daring to show it, and it reads honest performance numbers so it never puts something heavy on a struggling machine. You stay the director; it's the fastest pair of hands you've ever worked with.
.mcpb bundle). Double-click installs it into Claude Desktop. Other MCP-capable assistants point at the same connector file.The app must be running for the assistant to reach it. If the connection drops, the assistant says so plainly — nothing fails silently.
Everything below is a real capability, listed under the name the assistant will use when it narrates its work. The "say this" lines are starters — copy one into Claude and edit.
These are the assistant's eyes. It never has to guess what's on screen.
One call returns everything: the current scene and every slider, the saved-look list, whether a show is playing, the live music (tempo, key, energy), and honest performance numbers.
A good assistant calls this first, before touching anything — like a stagehand looking at the desk before moving a fader.
What's playing right now, and how does the look sound to you?
Returns the actual current frame as an image — what the audience sees this second. The assistant uses it to check its own changes.
Show me what the output looks like right now.
The killer feature: the assistant renders any scene offline at four musical moments — quiet, groove, build-up, drop — and gets back a four-panel contact sheet. It can judge a look the way a curator would, without ever putting a bad frame on your wall.
This is how the AI iterates safely during a live night: audition in private, show only when it's good.
Audition Kaleido before you switch to it — I don't want surprises.
The scene menu, readable by the assistant — including any scenes it has authored itself, which it can re-open and refine.
The app's built-in manual for making new scenes: what the music inputs are, the visual conventions, worked examples. Assistants read this before creating anything, so new scenes feel native.
Everything you can do on the panel, the assistant can do by name.
Change the visual world instantly (with your crossfade). Case doesn't matter — "kaleido" works.
Put Dunes on, slow crossfade.
Every slider from the control room by name: intensity, speed, warp, glow, trails, punch, ambient, brightness, the full color grade (exposure, contrast, lift, gamma, gain), the film pack (grain, halation, vignette…), and more. Values are clamped to safe ranges — an assistant cannot push anything somewhere dangerous.
More punch, a touch less glow, and warm the grade up slightly.
Pick one of the six palettes, apply a film LUT (Noir, Teal Orange, Warm Film, or your own), and flip the reactivity modes — macro reactive, key color, timbre glow.
Make the colors follow the song's key, and give it the Noir treatment.
The same curated one-press variation you have on the panel — matched to the music's current energy.
Surprise me, something that fits this track.
Set or clear the one-line text overlay — artist name, dedication, "happy birthday maya".
Put "LUNA — live" on screen in time for her set.
How long look changes take (hard cut to ten-second dissolve), and an optional frame-rate cap for quiet fans on long nights.
The two red buttons exist here too, so "kill it" works no matter who's driving.
Instant cut to black, fully restorable — the picture comes back exactly as it was.
Blackout until I say otherwise.
The guaranteed-sane reset: blackout cleared, full brightness, autopilot and show playback stopped, manual control returned. An assistant told "something's wrong, fix it" reaches for this.
Panic — give me manual control, everything visible.
The whole performance layer is scriptable by conversation.
List, apply, save, rename, reorder, delete — the entire look gallery. "Save this as Peak Hour" is one sentence.
Save the current look as "Golden Hour", and put it in slot one.
Build and edit the show timeline: add a cue from a look, set its hold time, reorder, rename, enable/disable, jump to any cue live.
Build a four-cue show: Dunes 2 minutes, Kaleido 90 seconds, Solar 2 minutes, end on Smoke.
Play, pause, loop, jump — the show's tape deck. Also reads back where the playhead is.
Start the show, loop it, and tell me when it wraps.
Turn hands-free mode on/off, set how often it changes (seconds or beats), how adventurous it is, and which looks it may draw from.
Auto-pilot every 64 beats, only from my saved looks, nothing too wild.
The deepest capability: the assistant can author brand-new visual worlds into your library.
From a description — yours in English, translated by the assistant into the app's scene language. It can build simple shader pieces, multi-layer scenes with persistent feedback, particle/fluid simulations with millions of particles, and even declarative 3D scenes with real materials and cameras.
New scenes appear in your menu alongside the built-ins and are saved with the app.
Make me a scene: a lighthouse beam sweeping through slow rain, mostly darkness, the beam widens on the drop.
The assistant iterates: check the scene compiles, audition it at the four musical moments, look at the sheet, adjust, repeat — the same taste loop a human designer runs, in seconds per lap.
The rain reads too busy — thin it out and make the quiet panels darker.
Remove an authored scene. Destructive, so assistants are asked to confirm before it happens. Built-in scenes can't be deleted.
Look at what's on screen. I'm playing slow house all evening — tune the look so it feels expensive: darker floor, warmer highlights, drops that swell instead of strobe. Check your work with a snapshot after each change.
Read the authoring guide, then build me a scene called "Harbor": navy darkness, a slow lighthouse beam, rain that thickens with the music's energy. Audition it at all four moments and iterate until the quiet panel is genuinely dark and the drop feels like the beam catching you. Then save a look for it called Harbor Night.
Build a 45-minute looping show from my saved looks — calm ones first, peak in the middle, land soft. 3-second crossfades. Start it, then check the performance numbers and tell me if anything's heavy.