Skip to the content
Music recommendation API

The music recommendation API that starts from an ISRC.

Send a track, an artist or a genre. Get back similar tracks, similar artists or a full playlist as ISRCs, ranked by score. Limit every answer to your own catalogue, and make it personal with your listeners' plays and skips.

~14M
tracks with a vector
721
genres to seed from
15.8M
playlists in the training data
similar-tracks · example values
curl 'https://api.musicapi.com/public/recommendations/similar-tracks?isrc=USUM71900764&limit=10' \
-u '<Client ID>:<Client Secret>'
# 200 OK · example values
{
"modelVersion": "20261114T020000Z",
"basis": "track",
"tracks": [
{
"isrc": "XXMAP2600001",
"title": "Example title",
"artist": { "id": "<artist id>", "name": "Example artist" },
"score": 0.9412
},
{
"isrc": "XXMAP2600002",
"title": "Another title",
"artist": { "id": "<artist id>", "name": "Another artist" },
"score": 0.9187
}
# … 8 more tracks
]
}
01 · How it works

Seed in, ranked ISRCs out.

We analyse no audio. The model learns which tracks listeners put on the same playlists, across music services, and we train it again every week.

01 · SEED

Send what you know

One ISRC, an artist id or a genre. A playlist takes up to 50 ISRCs, 50 artists and 20 genres. A negative weight steers away from a seed.

isrcartistIdgenreweight −10…10
02 · MODEL

A vector for each track

An item2vec model, trained on 15.8 million playlists, places tracks that listeners put together near each other. We train it again every week.

weekly retrainno audioplaylist co-occurrence
03 · ANSWER

Ranked ISRCs

Tracks ranked by score, up to 1. By default at most 2 tracks of one artist. Each answer names the model version that made it.

modelVersionbasismaxPerArtist
02 · Endpoints

Six jobs, one prefix.

Every route starts with /public/recommendations. Use your Client ID and Client Secret, as for every MusicAPI call. The routes that write or read listener data need the Client Secret.

Personal recommendations
Send play, complete, skip, like and dislike events with your own listener id. Recent events count more (half-life 30 days). A like counts most; a dislike pushes away.
Recommendation requests
GET/similar-tracksTracks that listeners put next to the seed ISRC.
GET/similar-artistsArtists like the seed, by artistId or by any ISRC of the artist.
POST/playlistA playlist of up to 200 tracks from weighted ISRC, artist and genre seeds.
GET/genresThe genres you can use as seeds, with their artist count.
Personal recommendations
GET/users/{userId}/recommendationsTracks for one listener, from the events you sent.
POST/nextThe next tracks in a session, from the recent and skipped ISRCs.
POST/eventsUp to 1,000 play, complete, skip, like and dislike events in one call.
DEL/users/{userId}Delete every event of one listener.
Catalogues
POST/cataloguesCreate a catalogue and get an upload URL for your ISRC file.
GET/cataloguesYour catalogues with their status and coverage.
03 · Catalogues

Recommend only what you can play.

Upload the ISRCs you hold the rights to. Add catalogue=<id> to a call and every answer comes from that list, with your own ISRC of each track.

▸A CSV or gzip CSV file with one ISRC on each line, up to 2 GiB
▸Upload it on the Recommendations page of the dashboard, or by API
▸Up to 10 catalogues for each account; add up to 10,000 ISRCs in one call
▸The first 100,000 ISRCs are free to host

Read the catalogue guide →

04 · The model

Built on what listeners put together, not on audio.

~14M
tracks with a vector
1.85M
artists
721
genres
15.8M
playlists in the training data
Weekly
a new model, trained again
The counts change with each weekly model.
05 · Pricing

Three meters. One invoice.

You pay for the calls you make, the listeners you personalise for and the ISRCs you host. Rates go down as volume goes up, within each calendar month.

Meter 1 · per 1,000 calls
€1.00/ 1,000 requests
/similar-tracks · /similar-artists · /playlist · /genres
up to 1M€1.00
1M – 10M€0.70
above 10M€0.40
Meter 2 · per listener a month
€0.04/ user / month
/events · /users/{userId}/recommendations · /next
up to 50,000€0.04
50,000 – 500,000€0.025
above 500,000€0.015
Meter 3 · per ISRC hosted
€10/ 1M ISRCs / month
/catalogues · catalogue=<id> on any call
first 100,000 ISRCsfree
every 1M after that€10

Recommendations have no minimum spend. The €500 minimum monthly spend applies only to standard requests with MusicAPI's managed credentials. Docs sandbox calls are free.

See full pricing →
06 · FAQ

Questions before your first call.

Do I need to upload a catalogue?

No. Without catalogue, answers come from the whole MusicAPI catalogue: about 14 million tracks with a vector. Upload a catalogue when you can play only your own licensed tracks. Then every answer comes from your ISRCs.

Which ISRCs work as a seed?

Any ISRC of the track. We accept dashes, spaces and lower case. When a track has no vector yet, the answer uses its artist's vector and says basis: "artist". /playlist returns the seeds that match nothing in unresolved.

How fresh are the recommendations?

We train the model again every week on new playlist data. Every answer carries modelVersion. We cache answers for one model version, so a new model gives new answers.

What personal data do you keep, and how do I delete it?

An event holds your own listener id, an ISRC, the event type and the time. We need no names and no email addresses. DELETE /public/recommendations/users/{userId} deletes every event of one listener, for example when they delete their account.

How do you meter it?

With three meters: recommendation requests per 1,000 calls, personalised users per listener a month, and catalogue hosting per million ISRCs. They go on the same monthly invoice as your transfer requests. Recommendations have no minimum spend. The €500 minimum monthly spend applies only to standard requests with MusicAPI's managed credentials.

Can I try it before I pay?

Yes. Sign in to the docs and call every read route from its docs page. Docs sandbox calls are free and do not count on your usage.

Get your first recommendations today.

Sign up, copy your Client ID and Client Secret, and call /similar-tracks. Or try every read route in the docs sandbox first.

Docs sandbox calls are free. Prices in EUR, excl. VAT.