spotify-mcp
MCP server for Spotify that enables searching, reading playlists, saved tracks, artist info, and recommendations, plus creating playlists and adding tracks, with no playback control.
README
spotify-mcp
An MCP server that wraps spotipy so Claude (or any MCP client) can search Spotify and read playlists, saved tracks, artist catalogs, and discovery signals.
Built specifically as re-com's Spotify backend — read-heavy by design, not a playback-control server. It does support creating playlists and adding tracks to them (so a re-com recommendation list can become a real playlist), but there's no play/pause/queue control; look at one of the several playback-focused Spotify MCP servers already out there for that.
Tools
| Tool | Description |
|---|---|
search_music(query, filter="track", limit=20) |
Search Spotify. filter is track or artist. |
get_playlists(limit=None) |
List the current user's playlists. Omit limit to fetch all of them. |
get_playlist_tracks(playlist_id, limit=None) |
Get the tracks in a playlist. Local files/episodes are skipped. |
get_saved_tracks(limit=None) |
Get the user's saved ("Liked Songs") tracks. |
get_track(track_id) |
Get a single track's metadata. |
get_recommendations(seed_track_id, limit=25) |
Spotify's algorithmic recommendations from one seed track — the closest analog to YouTube Music's radio. |
get_artist(artist_id) |
Get an artist's profile. |
get_artist_top_tracks(artist_id) |
An artist's top tracks (Spotify caps this at ~10 — there's no full-catalog endpoint). |
get_related_artists(artist_id) |
Artists related to the given one. |
get_recently_played(limit=50) |
The user's recently played tracks. |
create_playlist(name, public=False, description="") |
Create a new playlist owned by the current user. |
add_tracks_to_playlist(playlist_id, track_ids) |
Add tracks (by ID or URI) to a playlist, chunked in batches of 100. |
logout() |
Delete the cached OAuth token. |
A real limitation, stated plainly: Spotify restricts /recommendations and artist_related_artists for API apps created after November 2024 that don't have "Extended Quota Mode" (a manual approval Spotify grants sparingly). If your app doesn't have it, get_recommendations and get_related_artists will 403 — handle_errors turns that into a clear message rather than a raw traceback, and re-com's spotify_client.py treats it as one signal being unavailable, not a fatal error. search_music, playlists, saved tracks, and artist top tracks are unaffected.
Other Claude Code projects on this machine (e.g. re-com) call these tools by spawning this server over MCP rather than talking to spotipy/Spotify themselves — this is the only place Spotify credentials live.
Setup
1. Register a Spotify app
- Go to the Spotify Developer Dashboard, create an app.
- Add a redirect URI matching
SPOTIFY_REDIRECT_URI(defaulthttp://127.0.0.1:8888/callback— you don't need anything actually listening on that port; see step 3). - Note the app's Client ID and Client Secret.
2. Install dependencies
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
3. Authenticate
export SPOTIFY_CLIENT_ID="..."
export SPOTIFY_CLIENT_SECRET="..."
python scripts/setup_auth_spotify.py
This prints an authorization URL, waits for you to log in and paste back the URL you're redirected to (works fine over SSH/headless — nothing needs to bind the redirect port), and writes the resulting token to .spotify_cache (path configurable via SPOTIFY_CACHE_PATH).
.spotify_cache is equivalent to your logged-in session — never commit it or share it. It's already gitignored. Tokens refresh automatically once cached; re-run this script only if refresh itself starts failing (e.g. the app's client secret was rotated, or you revoked access from your Spotify account settings), or if SCOPE in server.py gains new permissions (delete .spotify_cache first so the auth flow re-prompts for consent — a stale cached token won't pick up new scopes on its own).
4. Add to Claude Code
claude mcp add spotify -s user \
-e SPOTIFY_CLIENT_ID="..." \
-e SPOTIFY_CLIENT_SECRET="..." \
-e SPOTIFY_CACHE_PATH="$(pwd)/.spotify_cache" \
-- "$(pwd)/.venv/bin/python" "$(pwd)/server.py"
-s user makes it available in any Claude Code session, not just this directory. Use absolute paths for the python interpreter, server.py, and SPOTIFY_CACHE_PATH since the server can be launched from any working directory.
For other MCP clients (Claude Desktop, etc.), point them at the same command and env vars using their respective config format.
Testing
The unit test suite (tests/) runs against a hand-rolled fake spotipy.Spotify client — no network access or Spotify credentials needed:
pip install -e ".[dev]"
pytest
Error handling
Tool calls translate common failure modes into clear messages instead of raw tracebacks:
- Missing/expired auth (401) → tells you to redo authenticate step.
- Restricted/forbidden (403) → tells you it's likely a Spotify API access restriction (see the recommendations/related-artists caveat above) or a missing OAuth scope.
- Rate limiting (429) → tells you to wait, including the
Retry-Afterhint if Spotify sent one. - Any other API or OAuth error is reported directly rather than as a raw traceback.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.