Tools

Ten read-only tools mapping the three MusicBrainz access modes — search when you have text, lookup (get_* and lookup_identifier) when you hold an MBID or standard identifier, and browse when you need the complete linked set beyond the single page that lookup folds in:

musicbrainz_search_entities

Resolve a name to an MBID with full-text Lucene search.

• Searches one entity type per call: artist, release-group, release, recording, work, or label
• Field-scoped Lucene syntax (e.g. artist:radiohead AND country:GB)
• Surfaces the raw 0–100 relevance score per hit (100 = exact); results stay in MusicBrainz score-descending order, not re-ranked
• Type-specific fields appear only for the relevant entity (ISRCs on recordings, ISWCs on works, artist credit on release-groups/releases/recordings)
• Pagination via limit (1–100) and offset; echoes the effective query and the true upstream total

musicbrainz_get_artist

The 80% artist-detail call.

• Folds discography (release-groups), band-membership / collaboration relationships, aliases, and tags/genres into one request via inc
• External links (Wikidata QID, Discogs, official site) surface as url-rels — chainable to wikidata-mcp-server and friends; this server does not chase them itself
• inc_release_groups and inc_relationships toggle the expensive sub-resources
• Discography and relationships are capped at one page (25); for a prolific artist's complete release-group list, use musicbrainz_browse_entities (target_type=release-group, artist link)

musicbrainz_get_release

One edition's full detail, the level with an actual tracklist.

• Tracklist as media → tracks → recordings, each with length and recording MBID (lengths rendered m:ss, stored as milliseconds upstream)
• Label + catalog number, barcode, country, release date, format, packaging, and text representation (language/script)
• Carries a cover-art availability stub from the WS/2 payload; call musicbrainz_get_cover_art with the release MBID for the actual image URLs

musicbrainz_lookup_identifier

The deterministic path when you already hold a standard identifier — no name search.

• id_type=isrc → recordings (a recording-level code, often shared by several recordings)
• id_type=iswc → works (a composition-level code)
• id_type=barcode → releases (UPC/EAN)
• ISRC and ISWC hit dedicated exact endpoints; barcode is a Lucene search filter, so its results are ranked (exact match scores 100)
• The output kind field tells you which entity type came back

musicbrainz_browse_entities

The complete-enumeration path — a correctness tool, not just convenience.

• Paginates the full linked set: every release-group by an artist, every release on a label, every recording of a work, every release in a release-group
• Pages arbitrarily deep via offset; totalCount is the true upstream total
• Use it whenever a linked set may exceed a page — the get_* tools embed at most one page (25), and a partial list read as complete is a silent correctness gap
• Provide exactly one link MBID matching a valid parent→child relationship for the target_type

musicbrainz_get_cover_art

Cover Art Archive images, kept separate from the release record.

• Front/back flags, image types, full-resolution URLs, and 250/500/1200px thumbnail URLs
• Returns an empty image set (not an error) when the entity has no art — absence of art is information
• Art is served at the release level; a release-group MBID resolves to a representative release's art automatically
• Image URLs are linked, never rehosted — image copyright stays with the rights holders (only the MusicBrainz core metadata is CC0)

Resources

All entity data is also reachable via the get_* tools, so tool-only clients (the majority) lose nothing. There is no resource list() — the corpus is millions of entities; discovery is via musicbrainz_search_entities, not resource enumeration.

Features

Built on @cyanheads/mcp-ts-core:

• Declarative tool and resource definitions — single file per primitive, framework handles registration and validation
• Unified error handling — handlers throw, framework catches, classifies, and formats
• Pluggable auth: none, jwt, oauth
• Swappable storage backends: in-memory, filesystem, Supabase, Cloudflare KV/R2/D1
• Structured logging with optional OpenTelemetry tracing
• STDIO and Streamable HTTP transports

MusicBrainz-specific:

• Type-safe client over the MusicBrainz Web Service v2 (musicbrainz.org/ws/2, JSON) plus the Cover Art Archive
• Process-wide ~1 req/sec token-bucket rate limiter — concurrent requests serialize to stay under MusicBrainz's per-IP ceiling, so multi-tenant load shares one budget
• Response caching keyed on the full request (including the inc set) — MBIDs are stable and entity data changes slowly, keeping most repeat lookups off the wire
• inc-driven lookups fold discography, relationships, tracklists, and external IDs into a single call rather than serial requests
• Retry with backoff over the full fetch + parse pipeline; an HTML error page served under load is classified transient, not as a parse error

Agent-friendly output: