Spotify OAuth 2.0 integration with Model Context Protocol (MCP) using Cloudflare Workers
The Spotify MCP server has 17 tools with consistent naming patterns (verb_noun structure) and basic descriptions. However, there are significant gaps in parameter descriptions, missing output schema documentation, and no error handling guidance. Most tools follow a simple search/read/write pattern with adequate naming but insufficient documentation depth for LLM-driven invocation. Parameter validation constraints are present (enums, defaults, limits) but descriptions lack actionable guidance on when to use each tool vs. similar ones.
Add tracks to a playlist
Create a new playlist
Get information about the user's current playback
Get the current user's Spotify profile
Get tracks from a playlist
Get the user's recently played tracks
Get the user's top artists
Output schemas not documented. Tool descriptions state success but do not specify what fields are returned, their types, or structure. LLMs cannot plan downstream operations without knowing response shape.
Error handling entirely absent. No guidance on what errors can occur, whether they are retryable, or what the LLM should do next. Tools return success text but no error recovery path.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 61 | <=2025-11-25 | v2 |
| 2026-03-09 | C | 61 | - | v1 |
Get the user's top tracks
Get the current user's playlists
Pause the user's playback
Resume the user's playback
Search for albums on Spotify
Search for artists on Spotify
Search for playlists on Spotify
Search for tracks on Spotify
Skip to the next track
Skip to the previous track
Destructive tools (pausePlayback, resumePlayback, skipToNext, skipToPrevious, createPlaylist, addTracksToPlaylist) lack confirmation mechanisms. Agents cannot preview impact before committing irreversible changes (e.g. skipping track, creating playlist).
Descriptions lack distinction between similar tools. searchTracks, searchArtists, searchAlbums, searchPlaylists all have identical parameter descriptions (query, limit). LLMs cannot infer when to use search_tracks vs search_albums without additional context.
No distinction in descriptions between read-only and write operations. pausePlayback and pausePlayback both state 'Pause the user's playback' with no indication this is a state mutation. LLMs need explicit state-change warnings.
Parameter descriptions generic and lacking context. 'Maximum number of results (1-50)' does not explain why limit matters, when to use small vs large limits, or what happens if results exceed limit. Descriptions should guide LLM decision-making.
No pagination details for list tools (getUserPlaylists, getPlaylistTracks, getRecentlyPlayed). Tools accept limit and offset but descriptions do not explain whether total count is returned, whether results are ordered, or whether offset is position-based or cursor-based.