# Spotify Skill > Read Spotify liked songs, playlists, top tracks and recently played, and export them as a DJ crate for Beatport matching. Use when Idan asks what he saved or liked on Spotify, wants to see or dump a playlist, or wants to turn earmarked tracks into a buy list. Read-only - it never modifies playlists or controls playback. - Skill: `idanbeck/spotify-skill` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add idanbeck/spotify-skill` - Raw SKILL.md: https://api.skillmd.com/api/skills/idanbeck/spotify-skill/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: idanbeck (https://skillmd.com/u/idanbeck) - Updated: 2026-09-09 - Page: https://skillmd.com/skills/idanbeck/spotify-skill --- # Spotify (read-only library + crate export) Pull liked songs, playlists, and top tracks out of Spotify, and export them in the shared crate format that `beatport-skill` matches against. **Read-only by design.** The requested scopes cover reading the library only — no playlist-modify, no playback control. Don't add write scopes without asking. ## The one thing to know: no BPM or key Spotify **permanently disabled** `audio-features`, `audio-analysis`, `recommendations`, `related-artists`, and 30-second previews for any app created after **2024-11-27**. A new app gets `403` on those endpoints — this is policy, not a bug, and no amount of scope fixes it. So a crate exported here has `bpm: null` and `key: null` **by design**. BPM and key come from Beatport during `beatport_skill.py match`. If Idan asks for the BPM of something he liked, that is the path — don't go looking for a Spotify endpoint that will serve it. What Spotify *does* still give, and what makes this whole workflow work: **ISRC**, on `track.external_ids.isrc`. That is an exact identifier, and matching a crate to Beatport by ISRC is exact rather than fuzzy. Always prefer a Spotify-sourced crate over a YouTube one. ## Setup Two ways in. **(b) needs no app at all** — use it if the developer dashboard refuses to create one, which it does fairly often with a generic "Something went wrong". ### (a) CSV import — the one that works right now **Spotify froze new developer-app creation in Dec 2025** and it is still frozen. The dashboard shows a Create app button that fails with a generic *"Something went wrong, we were not able to create your app"*, or says outright *"New integrations are currently on hold"*. Nothing about the form is wrong and retrying will not help. Route around it with an existing exporter — those authorize against apps registered long before the freeze, so they still work: 1. (open source, browser-only, read-only) - log in, export the playlist or Liked Songs to CSV. 2. Import it: ```bash python3 ~/.claude/skills/spotify-skill/spotify_skill.py import-csv ~/Downloads/friday.csv --name friday ``` Exportify's CSV carries **ISRC**, which is the field that makes Beatport matching exact — so this path loses nothing that matters. The importer maps columns by name and handles other exporters too (semicolon-delimited, `mm:ss` durations, missing ISRC); it reports which columns it used and the ISRC coverage. ### (b) Pasted web-player token — works, but off the sanctioned path ⚠️ **Tell Idan the tradeoff before he does this.** Spotify's own API response calls usage of the web player's token endpoint "not permitted under the Spotify Developer Terms and Developer Policy". Reusing your own session token is his call on his own account, but it is not a sanctioned integration path — prefer (a). Note the web player no longer calls `api.spotify.com` much, so filtering on that shows an empty Network tab. Filter on **`spotify.com`** and look for `api-partner.spotify.com` or `spclient.spotify.com` rows instead, then copy the `Authorization` header after `Bearer `: ```bash python3 ~/.claude/skills/spotify-skill/spotify_skill.py setup --token 'BQ...' ``` The web player's own token already carries the library scopes, so liked songs and playlists work immediately. **It expires after roughly an hour** and there is no refresh — on a 401 the skill says exactly how to re-copy it. Good for a one-off crate export; annoying if you're going to do this weekly. ### (c) Registered app + PKCE — best, if the freeze ever lifts 1. Create an app at . 2. Add the redirect URI **`http://127.0.0.1:8899/callback`**. **It must be the literal `127.0.0.1`.** Spotify explicitly rejects `localhost`; loopback is the one case where plain `http` is allowed. Getting this wrong is the #1 setup failure. 3. Copy the Client ID (PKCE is used, so **no client secret is needed** — don't ask for one). ```bash python3 ~/.claude/skills/spotify-skill/spotify_skill.py setup --client-id YOUR_CLIENT_ID python3 ~/.claude/skills/spotify-skill/spotify_skill.py login ``` **If "Create app" fails** with *"Something went wrong, we were not able to create your app"*: it is a long-standing dashboard bug, not something wrong with the form. Reported causes are an ad-blocker eating the POST, the loopback `http://` redirect URI tripping dashboard-side validation, and Web API access wanting a Premium account. Try incognito with extensions off; or create the app with a placeholder `https://example.org/callback` and add the `127.0.0.1` URI afterwards in Edit Settings, which is a different validator. If it still refuses, **use (a)** — it needs no app and gets the same data. `login` opens a browser and runs Authorization Code + PKCE against a one-shot local callback server. The refresh token is stored in `tokens/` (gitignored, chmod 600) and refreshed automatically, so `login` is a one-time step. ## Commands ```bash S=~/.claude/skills/spotify-skill/spotify_skill.py python3 $S me python3 $S liked [--limit 50] [--all] python3 $S playlists [--all] python3 $S playlist PLAYLIST_ID_OR_URL [--all] # accepts id, URL, or spotify:playlist: URI python3 $S top-tracks [--range short|medium|long] python3 $S recently-played [--limit 50] python3 $S search "query" [--limit 20] python3 $S logout ``` `--all` follows pagination to the end; without it you get one page. For a real set, use `--all` — a 300-track playlist silently truncated to 50 is a bad way to find out. ### Export a crate ```bash python3 $S export-crate --name friday --liked --all python3 $S export-crate --name friday --playlist "https://open.spotify.com/playlist/37i9..." --all python3 $S export-crate --name friday --top ``` Writes `~/.claude/skills/beatport-skill/crates/.json` and reports `with_isrc` — the count of tracks carrying an ISRC, i.e. how many will match Beatport exactly. Then: ```bash python3 ~/.claude/skills/beatport-skill/beatport_skill.py match ~/.claude/skills/beatport-skill/crates/friday.json ``` See `beatport-skill` for the rest of the pipeline and the crate schema. ## Output All commands print JSON. Track objects are already in crate shape (`source`, `source_id`, `artist`, `title`, `mix`, `isrc`, `duration_ms`, `url`), so any listing can be fed onward without reshaping. `mix` is parsed out of the track title when Spotify has it in parentheses — e.g. "Grey (Tale Of Us Remix)" yields `mix: "Tale Of Us Remix"`. That field is what stops Beatport matching from picking the wrong remix, so don't strip it. ## Notes - Standard library only — no pip installs, no spotipy. - `config.json` and `tokens/` are gitignored. - **Label is not available.** Spotify's track object doesn't carry it; the crate leaves `label: null` and Beatport fills it in during `match`. - On `429`, the skill stops and says it was rate limited. Wait, don't retry in a loop. - Podcast episodes in a playlist are skipped, not half-parsed as tracks.