Python · MCP

audio-analysis-mcp

A Python MCP server providing audio analysis tools for AI-driven sound recreation. It imports, separates, analyzes, and compares audio so an agent can configure hardware synthesizers to match a target sound.

Demucs stems Basic Pitch transcription Spectral + ADSR
★ View on GitHub Read the docs
What it does

Hear a sound, break it down

A suite of MCP tools take a target track from import, through separation and analysis, to a side-by-side comparison with the synthesized result.

Import & render
import_audioaudio_renderaudio_list_devices
Stem separation

Split into vocals, drums, bass, other, guitar and piano with Demucs.

stem_separate
Spectrum & envelope

Mel spectrogram, spectral features, ADSR and modulation.

spectrum_analyze
Transcription

Polyphonic transcription via Basic Pitch — MIDI plus note-event JSON.

note_transcribe
Note isolation
note_triagenote_isolate
Compare

Target vs. synthesized — mel spectrogram distance and per-band energy.

audio_compare
Setup

Up and running

Requires Python 3.11 and uv. audio_render needs PortAudio; system capture uses BlackHole.

~/audio-analysis-mcp
# install
uv sync --dev

# macOS audio capture deps
brew install portaudio

# run the MCP server (spawned by your client over stdio)
uv run python -m audio_analysis_mcp
Where it fits

The ears of the pipeline

audio-analysis-mcp turns a target recording into the features the agent reasons about.

Audio→Analysis→Agent→MIDI→Synth