io.github.cyanheads/musicbrainz-mcp-server

@cyanheads/musicbrainz-mcp-server

Search artists, releases, recordings, works, and labels; traverse relationships; resolve ISRC/ISWC/barcode; fetch cover art via MCP. STDIO or Streamable HTTP.

10 Tools • 1 Resource

Public Hosted Server: https://musicbrainz.caseyjhand.com/mcp

Overview

Open music metadata from MusicBrainz and the Cover Art Archive: artists, releases, recordings, works, and labels. Search by name, look up by MBID, ISRC, ISWC, or barcode, page through complete discographies, and fetch cover art.

Tools

Resources

Every entity is also reachable through the get_* tools, so tool-only clients lose nothing. The resource has no list; find MBIDs with musicbrainz_search_entities.

Capability reference

musicbrainz_search_entities tool

• One entityType per call (artist, release-group, release, recording, work, label) and a Lucene query with field scoping (artist:radiohead AND country:GB); positive integer limit (default 25, values above 100 clamp to 100 with a notice), offset for paging
• Returns the effective offset and upstream totalCount. Continuation uses offset + rows returned, including short pages. An empty page at or beyond a positive total reports exhaustion; an unexpectedly empty page before the total reports the discrepancy without suggesting a nonadvancing offset
• Each hit carries mbid, name, and the raw 0–100 score in MusicBrainz order, plus type-specific fields such as artistCredit, isrcs, iswcs, and length. Release rows use status (Official, Bootleg, …); artist, release-group, work, and label subtypes use type. A whitespace-only query fails as blank_query

musicbrainz_get_artist tool

• mbid, plus inc_release_groups and inc_relationships (both default true) to drop the heavier sub-resources
• Type, gender, country, area, lifeSpan, aliases, and tags, plus releaseGroups (one page), relationships (band membership, collaborations), and externalLinks (Wikidata, Discogs, official site)

musicbrainz_get_release_group tool

• mbid only; returns primaryType, secondaryTypes, firstReleaseDate, artistCredit with artistCreditString, tags, and one page of releases (editions)
• coverArt (exists, count, front, back) comes from a live Cover Art Archive check; when that check fails the field is omitted with a notice, never reported as "no art"

musicbrainz_get_release tool

• mbid, media_offset (default 0), and media_limit (default 20, no maximum); returns ordered media with complete tracks per medium (position, title, length as m:ss, recordingId), labelInfo (label and catalogNumber), barcode, status, date, country, packaging, language/script, and releaseGroupId
• mediaTotal and trackTotal count all normalized media and track occurrences; upstream per-medium trackCount is preserved. mediaOffset identifies the window. When more media remain, truncated, shown, cap, and a notice give the next media_offset; empty end pages report exhaustion. The release resource retains the complete raw tracklist
• coverArt is the availability stub from the release record; call musicbrainz_get_cover_art for image URLs

musicbrainz_get_recording tool

• mbid, plus inc_relationships (default true) for performer, producer, and engineer credits, work links to the composition, and external links
• Returns length, isrcs, artistCredit, firstReleaseDate, one upstream page of releases (up to 25), relationships, and externalLinks
• releaseAppearanceSummary reports shown, cap: 25, and completeness: a short present page is complete, a full or absent page without an exact count is unknown, and an exact upstream count exceeding the returned page proves truncated. totalCount is omitted when unknown; no count request is added. Enumerate appearances with musicbrainz_browse_entities (target_type=release, link.recording=<mbid>, offset=0), because embedded and browse order may differ

musicbrainz_get_work tool

• mbid, plus inc_relationships (default true) for writer, composer, and lyricist credits, movements and arrangements, the recordings that perform it, and external links
• Returns type, languages, iswcs, aliases, and tags. Only the first 25 recording relationships are kept (other relationships and links are all kept); totalCount gives the full count, and truncated, shown, and cap appear when some were cut
• To list every recording of a work, browse recording by work, then call musicbrainz_get_recording for relationship attributes and dates
• Follow a relationships entry with targetType: "work" using its targetId in another musicbrainz_get_work call. direction distinguishes parent/child links; optional orderingKey preserves the upstream sequence key (including zero), and targetDisambiguation distinguishes arrangements. Relationships retain upstream array order, which need not match movement order

musicbrainz_get_label tool

• mbid only; returns type, country, area, lifeSpan, labelCode (the LC number without its prefix), aliases, tags, and externalLinks
• Releases are not embedded; list them with musicbrainz_browse_entities (target_type=release, link.label)

musicbrainz_lookup_identifier tool

• id_type isrc → recordings, iswc → works, barcode → releases; a barcode is UPC/EAN digits and returns up to 25 ranked matches, each with a score
• ISRC accepts compact GBAYE9701274 or printed GB-AYE-97-01274, ignoring case and surrounding whitespace. ISWC accepts T-010.140.236-1 or T0101402361, with optional group punctuation, ignoring whitespace and case. Results echo the canonical compact ISRC or T-DDD.DDD.DDD-C ISWC; malformed codes fail before a request
• result.kind (recordings | works | releases) names the arm that came back; failures are invalid_identifier or identifier_not_found

musicbrainz_browse_entities tool

• target_type plus exactly one parent MBID in link; releaseGroup/release_group alias release-group. Positive integer limit defaults to 25; values above 100 clamp to 100 with a notice. Set MUSICBRAINZ_CLAMP_OVERSIZED_LIMITS=off to enforce maximum 100 for both search and browse
• Reports upstream totalCount; while more remain, truncated is set and a nonempty page gives the next offset using actual rows returned. Empty pages distinguish zero linked entities, exhaustion, and an unexpected empty page before the total. Clamp and paging guidance share one notice. Release rows use status; other entities use type
• A missing, extra, or malformed link fails as invalid_link, an unknown parent as entity_not_found. For ordered tracks of an edition use musicbrainz_get_release

musicbrainz_get_cover_art tool

• mbid and entity_type: release (default) or release-group, which resolves to a representativeRelease; limit defaults to 25 and clamps above 100, and offset defaults to 0. These windows apply after the archive index is cached and are independent of the search/browse clamp setting
• images retain archive order with front/back, types, imageUrl, and 250/500/1200px thumbnails. totalCount is exact, offset identifies the window, and hasArt reflects the full archive even on an empty end page. A single notice combines any clamp, continuation, exhaustion, or no-art guidance; truncated, shown, and cap appear when more images remain
• An unknown MBID fails as entity_not_found; if MusicBrainz can't confirm the MBID exists, the empty set comes back with a notice saying so

musicbrainz://{entity_type}/{mbid} resource

• entity_type is artist, release-group, release, recording, work, or label; returns the raw MusicBrainz record as application/json
• Fetches the same default linked data as the matching get_* tool; failures are invalid_mbid and entity_not_found
• Work JSON includes raw work-to-work relationships and keeps the first 25 recording relationships; recordingRelationshipSummary gives the exact recording count and whether any were cut
• Recording JSON carries the same releaseAppearanceSummary as the recording tool without changing cached raw data. Release JSON retains every medium and track; the tool's media window does not apply to resources

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

MusicBrainz-specific:

• Keyless client for the MusicBrainz Web Service v2 and the Cover Art Archive. Every request carries the descriptive User-Agent MusicBrainz requires, musicbrainz-mcp-server/<version> ( <contact> ), with MUSICBRAINZ_CONTACT as the contact
• Entities are addressed by MBID (a hyphenated UUID); uppercase and surrounding whitespace are normalized. Malformed or all-zero IDs fail as invalid_mbid (invalid_link for browse); a well-formed unknown ID fails as entity_not_found
• MusicBrainz calls are held to ~1 request/second across all clients of an instance, so long browse runs take a while
• Lookups fold discography, relationships, tracklists, and external links into one request; responses are cached (24 h by default) and transient upstream errors are retried with backoff

Agent-friendly output:

• Artist release-groups, release-group editions, and recording release appearances embed one page (25); browse enumerates the full set. Recording appearances disclose completeness without a count request. Work recording relationships are capped at 25 with the exact count reported; labels embed no releases
• Search echoes the effective query; search and browse report the exact upstream total. Missing or invalid totals fail as malformed upstream responses instead of inventing a count
• Discriminated outputs: musicbrainz_lookup_identifier returns a kind-tagged union, so callers branch on data, not string parsing
• Raw upstream relevance score, not a derived confidence; missing upstream fields stay absent rather than invented

Getting started

Public Hosted Instance

A public instance is available at https://musicbrainz.caseyjhand.com/mcp — no installation required. Point any MCP client at it via Streamable HTTP:

{
  "mcpServers": {
    "musicbrainz-mcp-server": {
      "type": "streamable-http",
      "url": "https://musicbrainz.caseyjhand.com/mcp"
    }
  }
}

Self-Hosted / Local

Add the following to your MCP client configuration file.

{
  "mcpServers": {
    "musicbrainz-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/musicbrainz-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info",
        "MUSICBRAINZ_CONTACT": "you@example.com"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "musicbrainz-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/musicbrainz-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info",
        "MUSICBRAINZ_CONTACT": "you@example.com"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "musicbrainz-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "MCP_TRANSPORT_TYPE=stdio",
        "-e", "MUSICBRAINZ_CONTACT=you@example.com",
        "ghcr.io/cyanheads/musicbrainz-mcp-server:latest"
      ]
    }
  }
}

For Streamable HTTP, set the transport and start the server:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 MUSICBRAINZ_CONTACT=you@example.com bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.4.0 or higher (or Node.js v24+).
• No API key or account. The server ships a default contact, but set MUSICBRAINZ_CONTACT to your own email or URL for any shared or hosted deployment.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/musicbrainz-mcp-server.git

1. Navigate into the directory:

cd musicbrainz-mcp-server

1. Install dependencies:

bun install

1. Configure environment:

cp .env.example .env
# edit .env and set MUSICBRAINZ_CONTACT (optional but recommended)

Configuration

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run:

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:stdio
  # or
  bun run start:http

• Run checks and tests:

  bun run devcheck   # Lint, format, typecheck, security, packaging
  bun run test       # Vitest test suite
  bun run lint:mcp   # Validate MCP definitions against the linter rules

Docker

docker build -t musicbrainz-mcp-server .
docker run --rm -e MUSICBRAINZ_CONTACT=you@example.com -p 3010:3010 musicbrainz-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/musicbrainz-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md/AGENTS.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for request-scoped logging, ctx.state for tenant-scoped storage
• Register new tools and resources via the barrels in src/mcp-server/*/definitions/index.ts
• Route every upstream call through the services, never a direct fetch(), so the User-Agent, rate limiter, and cache apply
• Wrap external API data: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.

MusicBrainz core entity data is CC0. The server reads only that core metadata and never fetches annotation text, which carries a different license. Cover art comes from the Cover Art Archive, a joint project of MusicBrainz and the Internet Archive; image URLs are linked, never rehosted, and each image's copyright stays with its rights holders. Cite MusicBrainz and the Cover Art Archive in downstream use.