.codex-plugin— mcp server

<div align="center"> <h1>@cyanheads/openlibrary-mcp-server</h1> <p><b>Search books and authors, fetch editions, browse subjects, and resolve cover images from Open Library via MCP. STDIO or Streamable HTTP.</b> <div>10 Tools • 2 Catalog Resources • 1 Agent Skill</div> </p> </div>

<div align="center">

</div>

<div align="center">

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

</div>

Overview

Open Library's catalog of 20M+ books, editions, authors, and subjects, plus full-text search inside Internet Archive's scanned books. Search and browse, drill from a work into its editions or an author into their works, and construct cover image URLs.

Tools

Resources

Both resources mirror data also available via openlibrary_get_work and openlibrary_get_author — useful for clients that don't surface MCP resources.

Agent skill

openlibrary-guide covers edition selection, bibliography resolution, passage identification, and incomplete metadata. It is included in npm, Docker, and desktop bundles and served through MCP skill discovery and skill:// resources.

Capability reference

Paginated tools return at most 100 rows per call. Larger positive integer limits are clamped with a notice showing the applied limit and next offset. OPENLIBRARY_CLAMP_LIMITS=false restores schema rejection above 100. Text and structured output carry the same complete domain data, including arrays and numeric precision.

openlibrary_search_books _tool

• Free-text query with Solr field prefixes (title:, author:, subject:, publisher:, isbn:, language:) or dedicated filter parameters; 1–100 results per page (default 10), offset pagination
• sort: relevance (default), new, old, rating, editions
• language accepts a 3-letter MARC code or a translatable 2-letter ISO code; an untranslatable 2-letter code fails as unknown_language_code rather than being silently dropped
• include_availability adds live Internet Archive borrow/read status (~200ms latency), off by default; flags Open Library leaves out are omitted rather than reported as false, and availability: null means none was returned for the work
• Returns work-level records with edition counts, cover IDs, subjects, and Internet Archive identifiers
• Optional matching_edition and matching_edition_count preserve the Search API's selected edition and matching count. The selected edition is provider-ranked; it does not prove that every unfielded search term matches that edition
• invalid_search_query reports upstream HTTP 422 validation with recovery guidance; empty queries with specific filters remain supported

openlibrary_get_work _tool

• Fetch by Open Library Work ID (OL…W), optionally /works/-prefixed; anything else — an ISBN, an edition or author OLID — fails input validation before any request, naming openlibrary_get_edition (id_type: "isbn") as the route from an ISBN to its work
• Returns title, description, subjects (plus place/time/people breakdowns), cover IDs, and author IDs — no author names (use openlibrary_get_author or openlibrary_search_books)
• A merged work ID stays reachable — the redirect chain is followed to the canonical record, work_id reports the canonical ID, and an enrichment notice names both IDs when they differ
• not_found when the Work ID doesn't exist or its redirect chain reaches no work

openlibrary_get_editions _tool

• List editions of a work by Work ID (OL…W), optionally /works/-prefixed; 1–100 per page (default 10), offset pagination, with the requested offset echoed in the output
• Same work_id validation as openlibrary_get_work — an ISBN resolves to its work through openlibrary_get_edition (id_type: "isbn")
• Returns ISBN-10/13, publishers, languages, page count, cover IDs, and edition OLIDs (OL…M), plus physical_format and edition_name when recorded
• A merged work ID stays reachable — the editions of the canonical work come back under its ID in work_id, with an enrichment notice naming both IDs; a live work still costs one request
• not_found when the Work ID doesn't exist or its redirect chain reaches no work

openlibrary_get_edition _tool

• Accepts 1–1,000 input strings, deduplicates exact strings, and processes the first 50 distinct identifiers in one upstream batch; the remainder is returned as unprocessed for resubmission. Every identifier shares one id_type: isbn (10 or 13 digits, an ISBN-10 may end in an X check digit), oclc (numeric), lccn (unchecked), or olid (OL…M). With OPENLIBRARY_CLAMP_LIMITS=false, the input array is limited to 50 entries
• Partial success: identifiers that resolve return in editions (request order); the rest land in unresolved with invalid_identifier (malformed, never sent upstream) or not_found (well-formed, no record) — the call fails only when nothing resolves
• upstream_unavailable (retryable) when Open Library's batch lookup itself fails — an HTTP error status other than a rate limit, or an HTML page — so an upstream outage never reads as a missing edition
• Authors come inline — the edition's own credits, or ones marked source: "work" recovered from the parent work when the edition itself lists none; if those lookups fail, the batch still returns, and an enrichment notice names the editions whose authors are missing or shown by ID
• Returns ISBN-10/13, OCLC, LCCN, LC call numbers, publishers, all cataloged languages, page count, cover IDs, parent work ID, and an Internet Archive ebook_url when one exists. The singular language remains the first cataloged language; physical_format and edition_name preserve the catalog's wording when present
• If no processed identifier resolves, the existing typed error also carries every unprocessed identifier in its data and recovery hint

openlibrary_search_authors _tool

• Search by name — partial and alternate names match; 1–100 per page (default 10), offset pagination
• Returns Author ID (OL…A), alternate names, birth/death dates, top work, work count, top subjects, and average rating

openlibrary_get_author _tool

• Fetch by Author ID (OL…A); a leading /authors/ prefix is stripped
• Returns bio, birth/death dates, photo IDs, and linked identifiers (Wikidata, VIAF, ISNI, Goodreads, LibraryThing)
• A merged author ID stays reachable — the response is the canonical record, and an enrichment notice names the canonical ID when it differs from the one requested
• not_found when the Author ID doesn't exist, names a record that is not an author (a work or edition OLID), or its redirect chain reaches no author

openlibrary_get_author_works _tool

• List works by Author ID (OL…A); 1–100 per page (default 20), offset pagination, with the requested offset echoed in the output
• Returns title, first-publish date, cover IDs, and Work ID (OL…W) per work
• A merged author ID stays reachable — an enrichment notice names the canonical ID when it differs from the one requested
• not_found when the Author ID doesn't exist, names a record that is not an author (a work or edition OLID), or its redirect chain reaches no author

openlibrary_get_subject _tool

• Subject name is trimmed, lowercased, and normalized from whitespace to underscores; 1–100 per page (default 12), offset pagination, with the requested offset echoed in the output
• Blank input fails as invalid_subject before any request
• Returns canonical subject name, normalized subject key, total work count, and per-work author names, edition count, and cover ID
• Empty results carry a recovery notice suggesting a different word form, synonym, or broader term — subject tags are user-contributed and inconsistent

openlibrary_search_inside _tool

• Full-text search across Internet Archive's scanned book text — the only tool that answers "which book contains this passage?"; quote a phrase for an exact match, unquoted terms match independently
• Usually takes 10–30 s against the live index, far above the metadata tools — reach for it deliberately, not as a general book search
• 1–100 results per page (default 10), offset pagination; each result carries a relevance score comparable only within its own result set
• Results key on Internet Archive ia_identifier, not Open Library work IDs — match it against ia_identifiers from openlibrary_search_books to reach the catalogue record
• upstream_unavailable (retryable) when the index answers without a result set, so a failed search never reads as "no book contains this"

openlibrary_get_cover_url _tool

• Resolves a cover or author-photo URL from id (numeric), isbn (10 or 13 digits, an ISBN-10 may end in an X check digit), or olid (OL…M for target: "book", OL…A for target: "author"); size is S/M/L (default M)
• Identifiers are validated locally before any request — path separators, .., and control characters fail as invalid_identifier, and an author lookup by isbn fails as invalid_target
• The tool constructs the URL without fetching it; a returned URL does not verify that a cover exists
• Output URL is ready to embed directly as ![cover](url)

openlibrary://works/{work_id} _resource

• Same fields as openlibrary_get_work, as injectable application/json context for a conversation about a specific book
• work_id comes from openlibrary_search_books or openlibrary_get_author_works; a merged ID resolves to the canonical work, whose ID the returned work_id carries, and an ID that is not a work (an edition or author OLID) is not found

openlibrary://authors/{author_id} _resource

• Same fields as openlibrary_get_author, as injectable application/json context for a conversation about a specific author
• author_id comes from openlibrary_search_authors; a merged ID resolves to the canonical author, whose ID the returned author_id carries, and an ID that is not an author (a work or edition OLID) is not found

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.

Open Library-specific:

• Open Library Search, Search Inside (full-text), Books, Authors, and Subjects APIs, cover URL construction, and Internet Archive availability lookups
• Work → editions and author → works drill-down, with explicit OLID cross-links between tool outputs
• Configurable User-Agent header (OPENLIBRARY_USER_AGENT) identifying the server per Open Library's bot-blocking convention
• Batch edition resolution — up to 50 ISBN/OCLC/LCCN/OLID identifiers in one upstream call, with per-identifier partial-failure reporting
• Timeouts sized per endpoint class (10 s record lookups, 30 s searches, 45 s full-text) inside a 50 s retry budget, so a slow upstream is waited on and a failing one returns a classified error before a typical 60 s client timeout; gateway errors are not retried in a loop

Agent-friendly output:

• Recovery guidance on every empty result — echoes the search criteria and suggests how to broaden a query or which offset to retry
• Merged-record disclosure — openlibrary_get_author, openlibrary_get_author_works, openlibrary_get_work, and openlibrary_get_editions follow merge redirects and surface the canonical ID via an enrichment notice when a requested ID was merged
• Per-item partial failure — openlibrary_get_edition returns resolved editions alongside typed unresolved reasons instead of failing the whole batch
• Complete text and structured output, with clamp and recovery notices preserved on both surfaces

Getting started

Add the hosted server to Claude Code — no API key required:

claude mcp add --transport http openlibrary-mcp-server https://openlibrary.caseyjhand.com/mcp

Or run it locally:

claude mcp add openlibrary-mcp-server -- npx -y @cyanheads/openlibrary-mcp-server

For other clients, add this to the MCP config file:

{
  "mcpServers": {
    "openlibrary-mcp-server": {
      "command": "npx",
      "args": ["-y", "@cyanheads/openlibrary-mcp-server"]
    }
  }
}

For Docker or a source build, see Running the server.

Configuration

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

Running the server

Local development

Requires Bun v1.4.0 or higher (or Node.js ≥ 24.0.0).

• Clone and install:

  git clone https://github.com/cyanheads/openlibrary-mcp-server.git
  cd openlibrary-mcp-server && bun install
  cp .env.example .env   # optional overrides; no required variables

• Build and run the production version:

  # One-time build
  bun run rebuild

  # Run the built server
  bun run start:http
  # or
  bun run start:stdio

• Run checks and tests:

  bun run devcheck  # Lint, format, typecheck, security
  bun run test      # Vitest test suite
  bun run lint:mcp  # Validate MCP definitions against spec

Docker

docker run --rm -p 3010:3010 ghcr.io/cyanheads/openlibrary-mcp-server:latest

The image defaults to HTTP transport, stateless session mode, and logs to /var/log/openlibrary-mcp-server. To run it over stdio from an MCP client, use "command": "docker" with "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/openlibrary-mcp-server:latest"]. To build locally, run docker build -t openlibrary-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 for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for logging, ctx.state for storage
• Register new tools and resources in the createApp() arrays
• Wrap external API calls: 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

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.