# music-metadata-mcp

MCP server for the [FreqBlog Music API](https://freqblog.com/?utm_source=github&utm_medium=readme).

Lets Claude, Cursor, Windsurf, and any MCP-compatible AI assistant look up audio features, build harmonic playlists, fetch lyrics, render waveforms, and export DJ-ready files. For reliable, full-catalog results, identify tracks by **name** (+ optional artist) or **ISRC** — a track name alone works, no Spotify account or ISRC needed. MusicBrainz IDs are also accepted. A raw **Spotify track ID** works too — resolved from our Spotify-ID map or, on a miss, by matching the track's title — but a title several artists share is ambiguous and misses rather than guessing, so it is not a universal Spotify-ID reverse lookup: prefer name or ISRC.

## Tools (v2.16.4 — 23 total)

### Core lookup
| Tool | Description |
|------|-------------|
| `lookup_track` | BPM, key, mood, genre, danceability, energy and 30+ more — best by track name (+ optional artist) or ISRC, which search the full catalog and queue an on-demand fetch + analysis on a miss; also accepts a MusicBrainz ID, or a Spotify track ID (resolved from our ID map or by title match; ambiguous titles miss rather than guess — prefer name/ISRC). Covers 800k+ fully analysed tracks + 7.5M fallback via MusicBrainz/AcousticBrainz (`GET /cache/stats` returns the live count as `tracks_analyzed`). Features come from a track's commercial streaming preview audio, so a track with no streaming release we can reach (CD-only, unreleased, or off-streaming) returns a clear "not available" result rather than data — there's no audio to analyse |
| `search_tracks` | Full-text search across the catalog (FTS5-backed) |
| `bulk_lookup` | Look up up to 50 tracks (by name or ISRC) in one request |
| `find_tracks_by_bpm` | Find tracks within ±tolerance of a target BPM |
| `find_tracks_by_key` | Find tracks by key — Camelot (8A), Open Key (1m), or name (A-Minor) |

### Recommendations & DJ tools
| Tool | Description |
|------|-------------|
| `find_similar_tracks` | Cosine-similarity recommendation engine over the entire catalog |
| `get_recommendations` | Spotify `/v1/recommendations` replacement — blend up to 5 seed tracks (or seed by `track`+`artist` name), genre-aware ranking. Tune with `min`/`max` (hard filters) and `target` (preference) over tempo, energy, valence, danceability, acousticness, instrumentalness, liveness, speechiness, loudness, popularity, duration_ms. Costs 2 units |
| `get_related_artists` | Spotify related-artists replacement — derived artist graph from audio-feature similarity. Costs 2 units |
| `build_radio_playlist` | Harmonic + BPM-continuity DJ playlist from a seed track |
| `score_transition` | Score how well one track mixes into another (0-100): harmonic + octave-aware BPM + energy. Costs 1 unit |
| `suggest_next_track` | Ranked next-track picks for a seed, each with transition score + reason. Costs 3 units |
| `build_setlist` | Order a 2-100 track crate into a beat-matched energy arc (peak_time/warmup/cooldown/flat). Costs 5 units |
| `export_playlist` | Generate Rekordbox XML / M3U / cuesheet from a list of track ids |
| `country_chart` | Live national music chart for 45 countries (Apple Music RSS) |
| `harmonic_keys` | Camelot wheel adjacency — pure logic, no quota |

### Browse
| Tool | Description |
|------|-------------|
| `find_artist_tracks` | List every catalog track for an artist (paginated) |
| `list_genres` | Distinct genre tags with track counts |
| `tracks_in_genre` | List tracks tagged with a genre |

### Per-track extras
| Tool | Description |
|------|-------------|
| `tag_track` | Compact, honestly-labelled tag list (energy/danceability/valence/acousticness/instrumentalness + mood + genre) — a tag-shaped projection of `lookup_track`, every tag carrying its own confidence + provenance. Costs 1 unit |
| `track_embedding` | 18-dim numeric vector for ML / similarity / clustering |
| `track_artwork_url` | Resolved cover-art image URL (iTunes / Cover Art Archive) |
| `track_lyrics` | Synced + plain lyrics via the open LRClib dataset |
| `track_waveform_svg` | SVG waveform render of the track's 30-second preview |

## Quick Start

Get a free API key at [freqblog.com](https://freqblog.com/?utm_source=github&utm_medium=readme#pricing) (1,000 req/month free).

### Claude Desktop

Add to `~/.claude/claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "music-metadata": {
      "command": "npx",
      "args": ["music-metadata-mcp", "--api-key=fb_live_YOUR_KEY_HERE"]
    }
  }
}
```

### Cursor / Windsurf

Add to your MCP config (`.cursor/mcp.json` or `.windsurf/mcp.json`):

```json
{
  "mcpServers": {
    "music-metadata": {
      "command": "npx",
      "args": ["music-metadata-mcp", "--api-key=fb_live_YOUR_KEY_HERE"]
    }
  }
}
```

### Environment variable

```bash
export MUSIC_API_KEY=fb_live_YOUR_KEY_HERE
npx music-metadata-mcp
```

## Example prompts

Once connected, you can ask your AI:

- *"What's the BPM and key of Blinding Lights by The Weeknd?"*
- *"Find me 10 tracks in A-Minor around 128 BPM"*
- *"What's the mood and genre of Come to Daddy by Aphex Twin?"*
- *"Look up the audio features for these 5 tracks: ..."*
- *"How well does this track mix into that one?"* — score_transition gives a 0-100 transition score with a reason
- *"What should I play after this track?"* — suggest_next_track ranks the best next picks for a seed
- *"Recommend tracks like these three"* — get_recommendations blends up to 5 seed tracks (Spotify /recommendations replacement)
- *"Which artists are similar to Van Halen?"* — get_related_artists derives an artist graph (Spotify related-artists replacement)
- *"Order these 12 tracks into a peak-time set and export it for Rekordbox"* — build_setlist then export_playlist
- *"Tag this track and tell me how confident each tag is"* — tag_track returns a compact tag list where every tag is labelled with its confidence + provenance (measured / derived / model-estimated / catalog-genre)
- *"Get the audio features for ISRC USUM71900001"* — name or ISRC give the best coverage; a MusicBrainz recording ID also works, and a Spotify track ID resolves from our ID map or by matching its title, and misses when several artists share that title (if your own Spotify integration already gives you the ISRC, pass that instead — it's exact)

## API key tiers

| Plan | Price | Requests/month |
|------|-------|---------------|
| Free | £0 | 1,000 |
| Hobbyist | £9.99/mo | 15,000 |
| Starter | £39/mo | 150,000 |
| Professional | £129/mo | 750,000 |

[Get your key →](https://freqblog.com/?utm_source=github&utm_medium=readme#pricing)

## Links

- [API documentation](https://freqblog.com/?utm_source=github&utm_medium=readme)
- [OpenAPI / Swagger docs](https://api.freqblog.com/docs)
- [GitHub](https://github.com/stevebirring-star/music-metadata-mcp)