Open-source · GPL-3.0

Recreate any sound
on real synths

An open-source AI agent that listens to a song, reverse-engineers its keyboard sound, and dials it into your hardware synths over MIDI — built from composable MCP servers you can run, fork, and extend.

Explore the repos → ★ Star on GitHub
Pipeline illustration: audio flows through an analysis engine and a MIDI data stream into a synthesizer, producing a recreated sound.
How it works

From audio to your synth, automatically

An AI agent orchestrates audio analysis and MIDI control to match a target sound on real hardware.

1

Listen

Import a track and isolate the keyboard part.

2

Analyze

Stem-separate, transcribe, and profile the tone.

3

Reason

The agent picks a synth and matching patch.

4

Configure

It sets the synth's parameters over MIDI.

5

Compare

Render, A/B against the target, refine.

The collection

Five repos, one pipeline

Each piece is independent and useful on its own — an MCP server, an analysis toolkit, or the agent that ties them together.

TypeScript

keyboards-mcp

MCP server that drives MIDI keyboards — Nord Electro 5D, Roland JUNO-X, Prophet-6 — plus a mock-device runner.

Learn more → Python

audio-analysis-mcp

Stem separation, spectrum analysis, transcription, note isolation and comparison — exposed as MCP tools.

Learn more → TypeScript

sound-recreation-agent

The AI agent (Vercel AI SDK) that orchestrates the MCP servers to recreate a target sound end to end.

Learn more → Research

reverse-synth-research

Design notes for the ML analysis pipeline — amplitude/ADSR, modulation, and tone-generation experts.

Learn more → Tooling

macos-packager

Will bundle the runtime pieces into a single macOS app — a planned one-click bring-up of the whole stack.

Learn more → Open source

Get involved

Building this in the open. If you're into audio analysis, DSP, or ML — come say hi, share an idea, or pick up a research problem.

Join the discussion →