Summary
librosa's analysis primitives, reachable from a chat.
The download tools are what make this practical: point at a YouTube link or a direct URL, the file lands locally, and tempo, beats, chroma and MFCC run on it in the same conversation. Worth knowing that what comes back describes the signal, not the recording's identity — nothing here tells you what the song is called.
What it is
A librosa-backed analysis server. It loads audio from disk, from a URL, or from YouTube, then reports duration, tempo, the beat track, chroma and MFCC features for the loaded signal.
What you get
- Audio loaded from a path, with an optional offset and duration so you can analyse just a section
- The total duration of the loaded signal
- Tempo in beats per minute, estimated by librosa
- A beat track — where the beats fall, not only how fast they come
- Chroma CQT, the chromatic content used for key and harmony work
- MFCC, the spectral description used for timbre and similarity work
- A file pulled down from any URL, or from a YouTube link, ready for the analysis tools
Requirements
Python with librosa; the project installs from its package index and runs with uv. No account and no key — analysis happens locally and the only network calls are the downloads you ask for.
Setup effort
One command — uvx -n mcp-music-analysis