Spotify Running Playlist Maker

Create playlists of tracks with a given tempo and genre.

The Problem: Spotify has the songs... but keeps removing the tools to find them.

Maybe you are a bit like me, an avid runner and a music lover. On my runs, I get tired of listening to the same few tracks over and over again. Spotify has more music than you can imagine, so I want some variation — in fact, A LOT of variation. At the same time, the music should still be tuned to my running pace.

This is what I use it for: this graph (from Garmin Connect) plots cadence over a running workout run with a playlist of just the right BPM.

But how do you find the right music?

• Spotify's interface gives you no way to browse by tempo or genre with any real precision. (Around 2015–2018 the Spotify Running feature kinda did the job; and then they pulled the plug on it.)
• Sure, you can always grab someone else's public running playlist, but are their tastes as bad as your own? Is the BPM range just right? For me, that's just not an option.
• EveryNoise did a tremendous job mapping genres, but it is now frozen in time as of December 2023, sabotaged by the Spotify layoffs that ended its development.
• For filtering music on tempo, tools like Playlist Machinery are great — but they only work on playlists you already have, meaning music you have already found.
• Spotify's own developer API, which this tool originally relied on for both genre-based artist search and tempo (BPM) data, had key endpoints removed in November 2024.

Maybe you are not like me at all. Perhaps you just want to scour Spotify for the next great track to blend into a DJ set. Or you have some other use for a genre-and-tempo crawler. Either way, read on.

The Solution: the Spotify Running Playlist Maker

This Python program will discover artists and tracks on Spotify within a given genre, measure their tempo, and save the songs that match your tempo filter to your Spotify profile as one or more new playlists.

The solution worked great, until it didn't anymore.

In fact, it broke beyond repair, or so I thought. What changed since the original version:

Spotify removed two critical API features for third-party apps in November 2024:

Feature	Old approach	New approach
Genre → artist discovery	Spotify search(type=artist, genre:...)	Last.fm tag-based search — free, crowd-sourced, continuously updated
BPM / tempo per track	Spotify Audio Features API	Deezer 30-second preview clips + librosa local beat analysis

EveryNoise is also no longer used as a genre reference — it has been frozen since December 2023 and does not reflect new music or evolving scenes.

However, Claude Code to the rescue! A new approach works around the critical APIs from other sources.

How it works

The functional flow is as follows:

1. You configure all your settings in settings.toml and run the program.
2. Genre → artist discovery (via Last.fm):
   • Fetches the top artists for your chosen genre tag across multiple pages of Last.fm results (configurable depth). Earlier pages surface the well-known names; later pages reach the long-tail artists.
   • Expands the pool further via a similarity graph: the top results are used as seeds for Last.fm's artist.getSimilar endpoint, which surfaces genre-authentic artists regardless of popularity. You can also add your own seed artists in settings.
   • All discovered artist names are resolved to Spotify artist IDs via name search.
3. Album and track enumeration (via Spotify): for each artist, all albums and singles are fetched, then all tracks within them.
4. BPM measurement (via Deezer + librosa):
   • The app searches Deezer's public API for each track to obtain a 30-second preview MP3.
   • The clip is downloaded temporarily and analysed locally by librosa's beat tracker.
   • The temp file is deleted immediately after analysis.
   • Results are cached in a local bpm_cache.db file so tracks are never downloaded and analysed twice across runs.
5. Filtering: only tracks whose BPM falls within your configured range are kept. Optionally, tracks at double or half the target BPM are also included (useful for tracks where the beat tracker locks onto a different subdivision).
6. Playlist creation: matching tracks are saved to one or more new Spotify playlists. Tracks already present in your existing playlists can be excluded — either by name, or wholesale with ["*"].

Setup

1. Create a Spotify app

Go to developer.spotify.com/dashboard and create an app. This gives you a Client ID and Client Secret.

In your app settings, add the following Redirect URI:

http://127.0.0.1:8888/callback

2. Get a Last.fm API key

Register at last.fm/api/account/create to get a free API key. No review process — it is issued immediately.

3. Configure settings.toml

Copy your Spotify Client ID, Client Secret, and Last.fm API key into settings.toml. Each setting is documented inside the file.

4. Set up a virtual environment and install dependencies

python -m venv .venv
.venv\Scripts\pip install -r requirements.txt   # Windows
# or
.venv/bin/pip install -r requirements.txt       # macOS / Linux

Key dependencies: spotipy, librosa, requests, tomli.

How to run

.venv\Scripts\python main.py   # Windows
# or
.venv/bin/python main.py       # macOS / Linux

First run only: a browser window will open asking you to authorise the app with your Spotify account. After authorising, the browser will redirect to a localhost address that shows "This site can't be reached" — that is expected. Copy the full URL from the address bar and paste it into the terminal. The authorisation token is then cached locally and this step does not repeat.

Once the program completes, your new playlists appear in your Spotify account as regular playlists — you can modify, delete, download, or collaborate on them like any other.

Generated playlist names include:

• The genre tag you searched
• The tempo range
• The date the program was run
• A "Part XX" suffix if you set a maximum number of tracks per playlist and multiple playlists were needed

Settings reference

Setting	Description
genre_searchstring	The Last.fm genre tag to search, e.g. darkwave, coldwave, ebm
artist_searchstring	Optional narrowing filter on artist name
interactive_mode	If true, prompts you to approve each artist before scanning
bpm_floor / bpm_ceiling	Tempo range in BPM
allow_doubled_bpm	Also include tracks at 2× or ½× the target BPM (handles beat tracker octave errors)
playlists_to_exclude	Playlist names to exclude tracks from. Use ["*"] to exclude all existing playlists — only brand new tracks will be added
max_tracks_per_playlist	Maximum tracks per playlist (0 = unlimited)
max_tracks_to_save	Maximum tracks across all playlists (0 = unlimited)
lastfm.api_key	Your Last.fm API key
lastfm.tag_pages	How many pages of top artists to fetch per tag (50 per page). Higher = more long-tail artists
lastfm.seed_artists	Optional list of artists to use as additional seeds for the similarity graph
debug	Set to true for verbose output including out-of-range BPM values

Technical notes

• Genre discovery uses Last.fm's tag API with configurable pagination depth, combined with artist similarity graph traversal to surface long-tail artists.
• BPM analysis runs entirely on your machine using librosa. Audio is sourced from Deezer's public preview API (no API key required). Preview clips are 30 seconds at 128 kbps (~480 KB each) and are deleted immediately after analysis.
• BPM results are cached in bpm_cache.db (SQLite) to avoid re-downloading and re-analysing the same tracks on subsequent runs.
• Per-run deduplication and playlist tracking use a separate in-memory SQLite database.
• The Spotify OAuth token is cached in a .cache file after first authorisation.

Known limitations

Duplicate tracks

The same recording may exist on Spotify under different URIs. The app deduplicates by track name as a best effort, but minor differences in capitalisation or punctuation can cause the same song to appear twice.

Artists that appear on other artists' albums

The Spotify get-an-artists-albums endpoint only returns albums where the artist is the primary credited artist. Guest appearances on other people's albums are not captured.

Artists not found at all

If an artist has no genre tags on Last.fm, they will not appear in tag-based discovery. You can still reach them directly by leaving genre_searchstring as a broad tag or space and specifying the artist name in artist_searchstring.

BPM accuracy

Librosa analyses a 30-second preview clip, which may start at an intro, outro, or quiet passage not representative of the track's main groove. The allow_doubled_bpm setting catches the most common error (beat tracker locking onto half-time or double-time). Tracks with genuinely ambiguous or variable tempo may be measured incorrectly.

Tracks with no Deezer preview

A small number of tracks have no 30-second preview available on Deezer. These are skipped and logged. BPM analysis is not possible for them without an alternative audio source.

Reporting issues

File bugs or suggestions here, or send a pull request.

Thanks for giving it a try — enjoy your new running playlists.