music-metadata-mcp

MCP server for the FreqBlog Music API.

Lets Claude, Cursor, Windsurf, and any MCP-compatible AI assistant look up audio features, build harmonic playlists, check lyrics availability, 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 per request by reading the track's title from Spotify's public embed page and matching it against our catalogue (we store no Spotify IDs; a resolution is remembered in memory for at most 24h) — 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 (v2.17.12 — 23 total)

Core lookup

Tool	Description
lookup_track	BPM, key, mood, genre, danceability, energy and 20+ 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 per request by title match against our catalogue — no Spotify IDs are stored; 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)

About the values (2026-09-27): audio features are FreqBlog's own analysis and are not on Spotify's scale. energy (since 2026-10-12) and valence (since 2026-10-13) are models of how energetic / positive the preview sounds, both on our own 0–1 scales — derive thresholds from our values rather than reusing Spotify-tuned ones. acousticness and danceability come from the AcousticBrainz classifier where available, else a model that listens to the preview (built on PANNs embeddings, CC BY 4.0; rolling out since 2026-09-27). instrumentalness (since 2026-10-12) is a vocal-detection model's score for the 30-second preview, not the whole track. speechiness and liveness are currently null for catalogue tracks.

Recommendations & DJ tools

Tool	Description
find_similar_tracks	Recommendation engine over the entire catalog — PANNs audio-embedding similarity blended with the 18-feature cosine
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, loudness, duration_ms (popularity, liveness, speechiness and instrumentalness are retired and return 400). 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 and each genre's parent genres
tracks_in_genre	List tracks tagged with a genre; include_subgenres pulls in the whole family (e.g. latin + urbano latino, pop latino)

Per-track extras

Tool	Description
tag_track	Compact, honestly-labelled tag list (energy/danceability/valence/acousticness + 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 (catalogue artwork / Cover Art Archive). Costs 1 unit
track_lyrics	Whether LRClib has plain/synced lyrics, plus a link to fetch them there (no lyric text). Free
track_waveform_svg	SVG waveform render of the track's 30-second preview. Costs 1 unit

Quick Start

Get a free API key at freqblog.com (1,000 req/month free).

Claude Desktop

Add to ~/.claude/claude_desktop_config.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):

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

Environment variable

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 per request by matching its title (read from Spotify's public embed page) against our catalogue, 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 →

Links

• API documentation
• OpenAPI / Swagger docs
• GitHub