Lineupify

Say what you want to hear. Get a Spotify playlist in seconds, in your own account, that you can read, edit and trust.

Lineupify is an MCP server for Claude Desktop, Claude Code, Cursor and any other MCP host. Paste a festival poster, describe a mood, name an artist you like, point at a playlist, or blend two people's playlists. Lineupify finds the artists, picks their most popular songs, matches them to Spotify and builds a draft you can review before anything is published. You bring a free Spotify developer app (two minutes, no client secret). Everything else runs on your machine.

• npm: lineupify-mcp (Node.js 20 or newer)
• Guides: Spotify app setup · Hosts · Troubleshooting · Security

Why

Streaming apps are good at playing music and bad at letting you say what you want.

• The daily mixes are not what you are in the mood for. They are built from what you played, not from what you want to play next.
• Shuffle keeps serving the same songs and the same artists. You want new songs by the artists you already love, not the same twenty again.
• You can describe a feeling but cannot find the genre. "Rainy Sunday jazz for cooking" is not a category in any app.
• Adding songs to a playlist takes minutes of searching when it should take one sentence.
• A festival lineup is forty names and you know six. Preparing for it by hand takes an evening.
• Blend only blends profiles. You want to compare two playlists, see what they share and why, and get a mix both people will actually like.
• The algorithm cannot explain itself. You want to see which artist a song came from and why it was picked, then change it.
• Your playlist is on Spotify but your friend is not. You want the list as a file they can use anywhere.

Lineupify answers each of these with a draft you can see, a reason for every track, and a playlist that ends up in your own account.

flowchart LR
    subgraph give[What you give]
        A[Festival poster or lineup text]
        B[A description or a mood]
        C[An artist you like]
        D[A country or the charts]
        E[A playlist link or your library]
        F[Two or more people's playlists]
    end
    subgraph engine[Lineupify]
        G[Artists]
        H[Most popular songs per artist]
        I[Matched to Spotify by ISRC]
        J[Filters, dedupe, order]
        K[Draft you can read and edit]
    end
    L[(Your Spotify account)]
    M[CSV / M3U / Markdown]
    A --> G
    B --> G
    C --> G
    D --> G
    E --> G
    F --> G
    G --> H --> I --> J --> K
    K -->|create_playlist| L
    K -->|export_draft| M

Quick start

The one step nobody can skip is creating a Spotify app, because Spotify only lets new apps serve their owner. Everything after that is one command.

1. Create a Spotify app (2 minutes). Go to https://developer.spotify.com/dashboard, click Create app, pick Web API, set the Redirect URI to exactly http://127.0.0.1:8765/callback, save, and copy the Client ID. No client secret is needed. Full walkthrough: docs/setup-spotify.md.

2. Run the guided setup in a terminal (needs Node.js 20 or newer):

   Terminal

   Copy

   npx -y lineupify-mcp init
   

   It takes the Client ID, logs you in through your browser, adds Lineupify to Claude Desktop, Claude Code or Cursor (whichever it finds), and runs a health check. Prefer clicking? Claude Desktop users can instead download lineupify-<version>.mcpb from the releases page and double-click it (Node.js 20 or newer must still be installed).

   Or let your assistant do step 2. If your assistant can run terminal commands (Claude Code, Cursor, and similar), paste this, with your Client ID filled in:

   Set up the Lineupify MCP server for me. Run npx -y lineupify-mcp setup --client-id PASTE_CLIENT_ID_HERE, then npx -y lineupify-mcp auth and tell me when to approve the login in my browser, then npx -y lineupify-mcp install --claude-code (or --cursor / --claude-desktop for the host you are running in), then npx -y lineupify-mcp doctor and show me the result. Windows users: run npx through cmd /c.

   If the server is already added but not connected (the .mcpb route, or a hand-written config), paste this in any chat:

   Call Lineupify's status tool. If Spotify is not connected, call connect with my Client ID PASTE_CLIENT_ID_HERE, tell me to approve the login in the browser, then call status again to confirm.

3. Ask for a playlist. Any of these work:

   • "Make me a playlist for this lineup" (paste the poster text or attach the image)
   • "Rainy Sunday jazz for cooking, about an hour, nothing explicit"
   • "Artists like Khruangbin, two songs each"
   • "New songs from my favourite artists that I have not liked yet"
   • "What does my friend's playlist have in common with mine? Then make us a mix."

No Spotify, or no Spotify Premium? Deezer mode

Spotify only lets a new app serve its owner, and that owner needs Premium. If that is not you, skip steps 1 and 2 entirely: add Lineupify to your host (docs/hosts.md) and ask for a playlist. With no Spotify login, drafts build on Deezer automatically (or say "use Deezer"; the option is provider: "deezer"). Deezer's public API is keyless, so there is nothing to create, paste or approve.

What works in Deezer mode: every seed except your own Spotify taste, every filter, reading, analysing, comparing and merging Deezer playlists and drafts, editing, and all exports. What does not: publishing into an account. Deezer stopped issuing API credentials to new apps in 2025, so no tool can write to a Deezer account today. If Deezer reopens its API, publishing will be added.

Getting a Deezer draft into your Deezer (or Apple Music, YouTube Music) account takes one paste:

1. Ask for the export: "export this draft as links" (export_draft, format: "links"; "text" gives "Artist - Title" lines instead).
2. Open TuneMyMusic or Soundiiz, choose import from text (TuneMyMusic: Let's start → Upload text; Soundiiz: Import playlist → From text), and paste the list.
3. Pick Deezer as the destination and confirm. The tool logs into your Deezer account itself; nothing about your account ever passes through Lineupify.

Both tools have free tiers that cover a normal playlist.

Spotify without Premium: borrow an app

Spotify's Premium rule applies to the person who owns the developer app, not to everyone who uses it. An owner can add up to four other people by email under User Management in the dashboard (five users per app in total), and those people log in with their own accounts, free ones included. So if someone you know has Premium:

1. They create the app as in step 1 above and add your Spotify account email under User Management.
2. They give you the Client ID (it is not a secret; the secret is never used).
3. You run npx -y lineupify-mcp init with that Client ID and log in as yourself. Your tokens stay on your machine; the owner never sees your account.

Two limits: an app serves five people at most at the time of writing, and the owner's daily API quota is shared across all of them and all of the owner's apps. Handing the Client ID to someone who is not on the User Management list does nothing; their login fails with a 403. Fine for friends and family; not a way to serve strangers, which Spotify's terms also rule out. Details: docs/setup-spotify.md.

What a conversation looks like

A festival

You: Make me a playlist for this lineup.

Code

Copy

SUNFALL FESTIVAL 2026 · 14-16 AUGUST
FRED AGAIN..   CHARLI XCX
Jamie xx · Four Tet · Overmono
Yaeji  Nia Archives  Barry Can't Swim  DJ Fluffhead
TICKETS ON SALE NOW

The assistant calls parse_lineup, which returns nine artists with tiers and drops the dates and ticket line, then create_draft with lineup: "Sunfall 2026". The draft comes back within about 15 seconds; a big lineup keeps building in the background:

Draft d_7k2mq "Sunfall 2026 · Lineupify"  rev 1  status ready  spotify: Alex
Artists 9 (resolved 8 · unresolved 1)
Tracks 27 · 1h42m · explicit 6 · via isrc 25 / text 2 · sources dz 27 / lfm 0 / sp 0
Tiers headliner 2×5 · sub 3×3 · undercard 4×2 · order interleave · private
Not found on Deezer or Spotify (1): DJ Fluffhead
Next: get_draft view=tracks to review, get_draft view=unresolved for misses, edit_draft to change, create_playlist with confirm: true to publish.

You: Show me the tracks.

t_4k2p  #1   Fred again.. – Marea (we've lost dancing)  4:52  dz/isrc  2021
t_c81z  #2   Charli xcx – Von dutch  2:44  dz/isrc [E]  2024
t_m0r3  #3   Jamie xx – Gosh  4:50  dz/isrc  2015
t_9x1c  #4   Four Tet – Baby  3:47  dz/isrc  2020
…

You: Drop the Four Tet track "Baby", forget DJ Fluffhead, and publish it.

edit_draft removes the track and excludes the artist; create_playlist returns the link:

Created playlist "Sunfall 2026 · Lineupify" with 26 tracks (1h38m).
URL: https://open.spotify.com/playlist/3cEYpjA9oz9GiPac4AsH4n

Later edits go through edit_draft followed by update_playlist, which replaces the playlist contents in place.

A mood

You: Rainy Sunday jazz for cooking, about an hour, nothing explicit.

The assistant proposes fitting artists itself, adds a genre seed with the same words so the list is not only its own guess, and calls create_draft with maxDurationMin: 60 and excludeExplicit: true:

Draft d_r47h8 "Rainy Sunday jazz · Lineupify"  rev 1  status ready
Seed genre "rainy sunday jazz" → 7 artists (Deezer playlists: Jazz for a Rainy Sunday Morning, a jazz Sunday in the rain, …)
Filters: clean only
Artists 10 (resolved 10 · unresolved 0)
Tracks 10 · 52:10 · explicit 0 · via isrc 10 / text 0

Two people

You: Compare my "6626" playlist with my listening history, then make a mix we would both like without anything that is already on it.

compare_playlists explains the overlap in numbers the assistant turns into words:

Compared 2 sides: 6626 (64 tracks, 52 artists) · your listening history (0 tracks, 210 artists)
Shared by all — artists (18): Mitski, My Chemical Romance, Linkin Park, Olivia Rodrigo, …
6626 vs your listening history: artist overlap 7% (18 shared, 0 identical tracks)
Only in 6626 (34): Camila Cabello, Sara Kays, Prelow, …

Then create_draft with seeds: [{ type: "blend", sources: ["6626", "me"] }] and excludeTracksFrom: ["6626"]:

Seed blend 6626 + me → 10 artists (99 artists on 2+ sides, 10 of the 10 picked on every side)
Excluded tracks from: 6626 (64 tracks)
Tracks 10 · 33:57 · explicit 1

How it works

Every request becomes an artist list, and every artist list goes through the same pipeline. Nothing touches Spotify until you publish.

flowchart TD
    A[artists and/or seeds] --> B[Expand seeds in the background<br/>genre · similar_to · chart · country · playlist · taste · blend]
    B --> C[Resolve each artist<br/>Deezer → Last.fm → Spotify]
    C --> D[Ranked candidate songs<br/>lead tracks first, featured next, live/remix last]
    D --> E[Match to Spotify by ISRC<br/>text search as fallback]
    E --> F{Filters}
    F -->|yearRange · bpmRange · explicit · excludeTracksFrom| G[Dedupe by URI, ISRC and title+artist]
    G --> H[Per-tier counts and maxTracks cap]
    H --> I[skipCovers · maxDurationMin · order]
    I --> J[(Draft on disk<br/>revisions, undo)]
    J -->|get_draft / edit_draft| J
    J -->|create_playlist| K[Spotify playlist]
    J -->|export_draft| L[CSV · M3U · Markdown]

Why Deezer and Last.fm for ranking? Spotify no longer exposes top tracks, recommendations, related artists, genres or audio features to new apps. Deezer's public API is keyless and gives popularity, related artists, tempo and playlists; Last.fm (optional key) adds tags, similar artists and per-country charts. Spotify is where the playlist ends up, and ISRC codes make the match exact.

A draft is a small state machine, checkpointed to disk after every artist so a killed process resumes where it stopped:

stateDiagram-v2
    [*] --> building: create_draft
    building --> ready: all artists fetched
    building --> paused: quota / token / network
    building --> failed: nothing to build
    paused --> building: get_draft
    ready --> building: edit_draft asks for more tracks
    ready --> published: create_playlist
    published --> published: edit_draft + update_playlist

Tools

Errors come back as CODE: message plus a Fix: line; every code is listed in docs/troubleshooting.md.

edit_draft operations

While a draft is still building, only exclude_artist, set_artist_track_count, set_artist_source, filter and set_meta are accepted. Pass expectedRevision from the last get_draft so an edit never applies to a list you have not seen.

Seeds: playlists without typing artists

The engine only needs an artist list. A seed produces one for you, alone or alongside typed artists.

Each seed adds up to limit artists (default 30, max 100) at the tier you give (default flat, or undercard when the typed artists have tiers). Seeds expand in the background build; the summary shows what each produced and why one failed, and a failed seed never blocks the rest.

Recipes the assistant can run in one call:

• Describe it: artists it proposes + seeds: [{ type: "genre", value: "<the words>" }].
• More like this: seeds: [{ type: "similar_to", value: "<artist>" }], tracksPerArtist: 2.
• Songs like these, other artists only: seeds: [{ type: "similar_songs", value: "<song link>", songs: [...] }], excludeSeedArtists: true, excludeTracksFrom: ["library"].
• 90s hip hop: seeds: [{ type: "genre", value: "hip hop" }], yearRange: { from: 1990, to: 1999 }.
• Running: any seed + bpmRange: { min: 160, max: 180 }.
• Expand my playlist: expand_playlist (a playlist seed + excludeTracksFrom the same playlist).
• New songs from my favourites: refresh_taste (a taste seed + excludeTracksFrom: ["library"]).
• Blend for a road trip: compare_playlists to explain the overlap, then seeds: [{ type: "blend", sources: [...] }] with excludeTracksFrom the same sources so nothing anyone already has is repeated.
• Merge: merge_playlists keeps the actual tracks and drops duplicates.

read_playlist, analyze_playlist and compare_playlists return plain data lines (counts, decades, genres, tempo buckets, overlap percentages). The assistant turns them into words, tables or charts; the server never draws.

No Spotify account for the other person? export_draft gives Markdown, CSV or M3U they can take anywhere.

create_draft options

Every summary and publish result ends with the artists that were not found or had no playable Spotify track, and what to do about them. Defaults can be changed permanently with lineupify-mcp config set (see Configuration).

CLI reference

Running lineupify-mcp with no arguments serves MCP over stdio; that is what your host runs. The subcommands below are for a normal terminal (npx -y lineupify-mcp <command> or, after a global install, lineupify-mcp <command>).

Example doctor output:

OK   Node.js          22.11.0
OK   Data dir         /home/alex/.lineupify
OK   Client ID        set (config.json)
OK   Redirect URI     http://127.0.0.1:8765/callback (port free; this exact URI must be in the app's Redirect URIs)
OK   Spotify login    Alex; refresh token valid 171 more days
OK   Spotify API      GET /me ok (alexr)
OK   Deezer           HTTP 200
OK   Last.fm key      not set (optional)

Configuration

Settings are read from environment variables first, then from ~/.lineupify/config.json. An environment variable always wins over the file.

Environment variables

Pass them through your host's env block (see docs/hosts.md).

config.json

{
  "spotifyClientId": "0123456789abcdef0123456789abcdef",
  "spotifyRedirectPort": 8888,
  "lastfmApiKey": "…",
  "defaults": {
    "tracksPerTier": { "headliner": 5, "sub": 3, "undercard": 2 },
    "maxTracks": 250,
    "order": "interleave",
    "public": false,
    "excludeExplicit": false,
    "allowVersions": false,
    "discoveryOnly": false,
    "skipCovers": false,
    "namingTemplate": "{lineup} · Lineupify"
  }
}

spotifyRedirectPort and lastfmApiKey are optional. Everything under defaults is optional and overrides the built-in defaults listed in the options table.

config set keys

Example: lineupify-mcp config set tracksPerTier.headliner 8

Privacy and data

Lineupify runs on your machine with a Spotify app you created. It has no server of its own, no telemetry, and no access to your account beyond the token on your disk. SECURITY.md lists what it can and cannot do and how to report a problem.

What is stored, all under ~/.lineupify/ (or LINEUPIFY_HOME):

tokens.json is written with mode 0600 on macOS and Linux. On Windows that call is a no-op and the file is protected by your user profile's default permissions, the same as other CLIs' credential files.

Spotify permissions requested at login, and what needs each one:

Lineupify never deletes or unfollows a playlist, never changes your library or follows, and creates playlists private unless you ask for public. The only overwrite is update_playlist on a playlist Lineupify created, and it refuses if that playlist changed inside Spotify unless forced.

Where data goes. Lineupify talks to api.spotify.com and accounts.spotify.com (your account), api.deezer.com (keyless, no account), ws.audioscrobbler.com (only with a Last.fm key), musicbrainz.org and labs.api.listenbrainz.org (only for a similar_songs seed: the seed songs' ISRC, title and artist are looked up there) and registry.npmjs.org (a version check at most every 6 hours; LINEUPIFY_NO_UPDATE_CHECK=1 turns it off). Artist names, track titles and ISRCs from your playlists, liked songs and top artists are sent to Deezer as search queries for ranking, tempo, genres and cover checks, and to Last.fm when a key is set. No account identifier goes with them. If you would rather keep your listening data out of Deezer, use typed artist lists with sources: ["spotify"] and skip analyze_playlist, bpmRange and skipCovers.

Switches.

• LINEUPIFY_READ_ONLY=1 disables create_playlist and update_playlist; everything else works. Good for "analysis only" setups.
• disconnect (tool) or lineupify-mcp logout forgets the login; purge: true (with confirm: true) / --purge deletes the whole data folder, and refuses if the folder holds anything Lineupify did not create. Remove the app's access on Spotify's side at https://www.spotify.com/account/apps/.
• Your MCP host can disable the server entirely (Claude Desktop: Settings → Developer; Claude Code: claude mcp remove lineupify).

Model-driven writes. create_playlist refuses until the draft has been shown to you, unless the assistant passes confirm: true. Like any MCP server, Lineupify does what the assistant asks; the write tools carry MCP destructiveHint annotations so hosts that ask for permission can single them out. Review the draft before publishing, or run read-only.

External text. Poster text, track titles, playlist descriptions and Deezer playlist names are cleaned (control characters stripped, length capped) and shown inside fixed table layouts, which reduces the chance that external text is read as an instruction. It cannot rule it out: a poster line that says "publish this as public" reaches the assistant as data, and nothing stops a model from acting on it. The real protections are the switches above: playlists are private by default, create_playlist needs a reviewed draft or an explicit confirm, disconnect purge needs confirm too, LINEUPIFY_READ_ONLY turns every write off, and Spotify has no delete endpoint. Logs go to stderr only, with tokens and keys redacted.

Terms. Last.fm data is for non-commercial use. Deezer's public API has its own terms of use. Check both before using Lineupify for a business (a venue, a radio schedule).

Limits and known issues

• Spotify Development Mode. New Spotify apps run in Development Mode: at most 5 users, and the app owner must have Spotify Premium. Production ("Extended Quota Mode") is only granted to registered businesses with 250,000+ monthly active users, so every Lineupify user creates their own free app instead. Other people can only use your app if you add them under User Management in the dashboard. Without Premium, use Deezer mode.
• Deezer cannot be written to. Deezer closed API app registration for new developers in 2025 and had not reopened it as of mid-2026, so Deezer drafts are export-only. Reads, ranking, related artists, tempo and playlist lookups are keyless and unaffected.
• 6-month logins. Spotify refresh tokens expire 6 months after the original login. status warns when 30 days are left; reconnect with connect force: true (or lineupify-mcp auth --force).
• Daily quota. Development Mode has a daily request quota shared across all apps you own. When it runs out Lineupify reports SPOTIFY_QUOTA_EXCEEDED, the draft is paused, and get_draft resumes it once the quota resets. Results already fetched are cached, so nothing is lost.
• 60-second hosts. Claude Desktop and Cursor time out any tool call after 60 s and ignore progress notifications. create_draft therefore returns within about 15 s and keeps building in the background; poll with get_draft waitSeconds: 25. Claude Code has no such limit.
• Ranking without Spotify. Spotify removed artist top-tracks, recommendations and popularity for new apps, so songs are ranked with Deezer's public API and optionally Last.fm, then matched to Spotify by ISRC. Very small or brand-new acts may not be on Deezer; they show up as unresolved. Adding a Last.fm key helps; add_track covers the rest.
• Size. Up to 400 artists per draft (split bigger lineups by day; seeds fill the remaining room) and 250 tracks by default (maxTracks, up to 10,000).
• Reading playlists. Playlists made by Spotify itself (Discover Weekly, Blend, Today's Top Hits, Daily Mix) cannot be read by new apps; playlists made by people can, when public or in your own library. Reads are capped at 1,000 tracks (3,000 for liked songs). If status lists missing permissions after an upgrade, reconnect with connect force: true.
• Genres and tempo. Spotify gives new apps no genres or audio features, so analyze_playlist uses Deezer's coarse genres (Pop, Rock, Metal, …), Last.fm tags when a key is set, and Deezer tempo sampled over up to 60 tracks. Remastered releases carry the remaster year, so yearRange treats them as unknown unless strictYear is on.
• similar_songs without a Last.fm key uses ListenBrainz alone, whose coverage is thinner for small or non-Western artists; a key adds Last.fm's co-listening data. Both sources work at recording level, so the song must exist in MusicBrainz (ListenBrainz) or have Last.fm scrobbles.
• Seeds without a Last.fm key rely on public Deezer playlists for genre and country; results are good for common genres and large countries and thinner for niche tags. A Last.fm key (setup lastfmApiKey) adds tag, similar-artist and per-country data.
• One builder at a time. If two hosts (for example Claude Desktop and Claude Code) run Lineupify at once, only one of them builds a given draft; the other reads it.
• npx caching. npx -y lineupify-mcp keeps the first version it downloaded. Update with npm i -g lineupify-mcp@latest or npx -y lineupify-mcp@latest. status tells you when a newer version exists.

Development

npm install
npm run typecheck && npm run lint && npm test   # unit tests, offline (Spotify and Deezer mocked); includes a stdio boot test of the server
npm run coverage                                 # the same with a coverage report
npm run test:live                                # live Deezer checks
npm run smoke:spotify                            # every Spotify endpoint, needs a connected account
npx tsx test/smoke/playlists.ts <playlist link>  # reads, analysis and seeds against live APIs
npm run build                                    # dist/
npm run bundle:mcpb                              # build/lineupify-<version>.mcpb for Claude Desktop

Changes are listed in CHANGELOG.md.

Credits

• Song ranking, related artists, tempo and playlist data from the Deezer public API.
• Powered by Last.fm data when a LASTFM_API_KEY is configured. Last.fm data is for non-commercial use.
• Song similarity uses open data from MusicBrainz and ListenBrainz (MetaBrainz Foundation), no key needed.
• Playlists are created through the Spotify Web API.
• Built on the Model Context Protocol (@modelcontextprotocol/server).

Lineupify is not affiliated with Spotify, Deezer or Last.fm.

License

MIT. See LICENSE.