WebAudio · MCP · runs on your machine
The synth your AI plays.
Not a track it generates. Sixty-seven real parameters, turned live in this tab — and a score, stems and MIDI you keep.
- Free, and it runs locally
- The AI turns real knobs
- Score, stems and MIDI out
- Same seed, same audio
Click anywhere or press a key to start audio
Play with your mouse, or use keys A–; (black keys on W E T Y U…).
Model Context Protocol
Make music with an AI
WaveCraft ships an MCP server, so Claude — or any MCP client — sits down at this exact instrument. Not a prompt box wired to a model: the same 67 parameters, 46 presets and 24 drum voices you have, driven note by note and bar by bar.
With this tab open it all sounds live and you can jam over it. With no tab — a terminal, a sandbox, a server — the same tools render offline and hand back the audio. Either way you get a mixed WAV, separate stems, and a MIDI file.
Prefer to start from a sentence? Open the Studio — type a prompt, and the song comes back as stems on a mixing console, rendered right in your browser. Ride the faders, swap an instrument, or ask for changes; only what changed re-renders. Four sessions are already written and waiting there, so you can hear the whole thing before you sign up for anything.
The argument
Why an instrument, not a generator
If you've reached for a prompt-to-audio tool — ElevenLabs Music, Suno, Udio — you know the deal: you describe a song, wait, and get back a finished recording. It's remarkable, and it's the wrong shape for actually making something. WaveCraft goes the other way.
-
The score is the artifact
A song here is a JSON document — tracks, patterns, chord progressions, arrangement, automation. Read it, diff it, commit it, hand-edit one note. The audio is just what falls out when you render it.
-
Change one thing, change one thing
Renders are deterministic: same song, same seed, byte-identical audio. Drop the bass an octave in the chorus and that's all that moves. No re-roll, no losing the take you liked.
-
Sound design is a conversation
“Open the filter, more drive, longer release” isn't a hope you write into a prompt — it's 67 real parameters being turned. Ask why it sounds like that and the answer is a patch you can inspect.
-
MIDI out, straight to your DAW
Export the notes, not just the mix. Take the arrangement into Ableton or Logic and play it back through your own instruments, or your own hardware.
-
You're in the room
Keep this tab open and the AI's parts sound in real time, on the synth in front of you. Play over the chords, ride the cutoff mid-phrase, hit Panic. You're a player, not a reviewer.
-
Local, free, and yours
One
npxand it runs on your laptop. No account, no credits, no queue, nothing uploaded — the music never leaves your machine unless you send it somewhere.
| Prompt-to-audio generators | WaveCraft | |
|---|---|---|
| What comes back | A finished recording | A score you can edit — plus WAV, stems and MIDI |
| Making a small change | Ask again, take what you get | Edit the note, re-render; everything else is identical |
| Sound design | Described in words | 67 parameters, turned in real time |
| Where it runs | Their servers, metered | Your machine, free, offline |
| Your role | Reviewer | Player — jam along in the browser |
| Strongest at | Vocals, real instruments, finished tracks | Synths, electronic music, arrangements you'll keep working on |
To be straight about it: this is a synthesizer. There are no vocals and no sampled orchestras — if you need a singer, go use a generator. What you get instead is every note, every knob and every bar under your hands, and a collaborator who can actually reach them.
Sponsored