MCP Open Library

A Model Context Protocol (MCP) server for the Open Library API that enables AI assistants to search for book and author information.

Overview

This project implements an MCP server that provides tools for AI assistants to interact with the Open Library. It allows searching the catalogue by title, author, subject and other fields, searching for authors by name, retrieving detailed author information using their Open Library key, and getting URLs for book covers and author photos. The server returns JSON projections of the Open Library responses rather than the raw payloads.

Features

• Book Search: Search across titles, authors, subjects, places, people, publishers and ISBNs, with sorting and paging (search_books).
• Book Search by Title: Search for books using their title (get_book_by_title).
• Author Search by Name: Search for authors using their name, with paging (get_authors_by_name).
• Get Author Details: Retrieve detailed information for a specific author using their Open Library key (get_author_info).
• Get Author Photo: Get the URL for an author's photo using their Open Library ID (OLID) (get_author_photo).
• Get Book Cover: Get the URL for a book's cover image using various identifiers (ISBN, OCLC, LCCN, OLID, ID) (get_book_cover).
• Get Book by ID: Retrieve detailed book information using various identifiers (ISBN, LCCN, OCLC, OLID) (get_book_by_id).

Search results are paged — every search tool returns at most limit results (default 10, maximum 50) alongside num_found, the total number of matches, which you page through with offset (maximum 1000). The two cover tools check that an image actually exists and say so when it does not, rather than handing back a URL that resolves to a blank placeholder.

Every tool is a read-only lookup and advertises itself as such with the readOnlyHint and openWorldHint annotations, which may allow a client to skip the confirmation prompt it shows for tools that could change something. These are hints: the MCP specification has clients treat annotations as untrusted unless the server is trusted, so the confirmation policy is the client's to decide. Failures — an unreachable API, a rejected argument — come back as a tool result flagged isError, so an assistant can read what went wrong and correct its next call rather than the request failing outright.

Installation

Quick Start

Nothing to install or build. Point an MCP client at the package with npx and it will be fetched on first run:

{
  "mcpServers": {
    "mcp-open-library": {
      "command": "npx",
      "args": ["-y", "mcp-open-library"]
    }
  }
}

In Claude Desktop that goes in claude_desktop_config.json; other clients use the same shape. Restart the client and the seven tools below become available.

MCP Registry

This server publishes to the official MCP Registry as io.github.8enSmith/mcp-open-library from v1.0.3 onwards. Clients that support the registry can install it by that name.

To inspect the published listing:

curl "https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.8enSmith/mcp-open-library"

Manual Installation

# Clone the repository
git clone https://github.com/8enSmith/mcp-open-library.git
cd mcp-open-library

# Install dependencies
npm install

# Build the project
npm run build

Usage

Running the Server

1. Ensure you are running node v22.21.1 (it'll probably work on a newer version of node but this is what Im using for this test). If you have nvm installed run nvm use.
2. In the mcp-open-library root directory run npm run build
3. Next run npm run inspector. Once built, click the URL with the MCP_PROXY_AUTH_TOKEN query string parameter to open the Inspector.
4. In the Inspector, choose 'STDIO' transport
5. Make sure the command is set to 'build/index.js'
6. Click the 'Connect' button in the Inspector - you'll now connect to the server
7. Click 'Tools' in the top right menu bar
8. Try running a tool e.g. click get_book_by_title
9. Search for a book e.g. In the title box enter 'The Hobbit' and then click 'Run Tool'. Server will then return book details.

Using with an MCP Client

This server implements the Model Context Protocol, which means it can be used by any MCP-compatible AI assistant or client e.g. Claude Desktop. The server exposes the following tools:

• search_books: Search the catalogue by any combination of query, title, author, subject, place, person, publisher and ISBN
• get_book_by_title: Search for book information by title
• get_authors_by_name: Search for author information by name
• get_author_info: Get detailed information for a specific author using their Open Library Author Key
• get_author_photo: Get the URL for an author's photo using their Open Library Author ID (OLID)
• get_book_cover: Get the URL for a book's cover image using a specific identifier (ISBN, OCLC, LCCN, OLID, or ID)
• get_book_by_id: Get detailed book information using a specific identifier (ISBN, LCCN, OCLC, or OLID)

Example search_books input:

{
  "author": "Ursula K. Le Guin",
  "subject": "fantasy",
  "sort": "old",
  "limit": 2
}

Example search_books output:

{
  "num_found": 51,
  "offset": 0,
  "limit": 2,
  "results": [
    {
      "title": "A Wizard of Earthsea",
      "authors": ["Ursula K. Le Guin"],
      "first_publish_year": 1968,
      "open_library_work_key": "/works/OL59798W",
      "edition_count": 87,
      "author_keys": ["OL31353A"],
      "best_edition": {
        "edition_key": "OL5613890M"
      },
      "cover_url": "https://covers.openlibrary.org/b/id/13617691-M.jpg",
      "ratings_average": 3.95,
      "ebook_access": "borrowable"
    }
  ]
}

best_edition is one specific edition of the work — the one Open Library ranks best for your query — carrying that edition's own identifiers. Search results otherwise identify a work (open_library_work_key), which no tool accepts, so this is the route from a search hit to a concrete book.

Its edition_key is an OLID you can pass straight to get_book_by_id for the full edition record, including its complete ISBN arrays:

{ "idType": "olid", "idValue": "OL5613890M" }

The isbn_13 / isbn_10 fields are omitted where Open Library holds no ISBN for that edition — as in the example above, and roughly a third of results — while edition_key is essentially always present. Where an edition lists several ISBNs of one kind, the first is reported; get_book_by_id returns them all.

The search_books tool accepts the following parameters:

• At least one of q, title, author, subject, place, person, publisher or isbn — the request is rejected without one, since an unfiltered search matches the entire catalogue. q takes a free-form Solr query such as subject:cyberpunk AND first_publish_year:[1980 TO 1990]
• language: Optional 3-letter MARC language code (e.g. eng, fre)
• sort: Optional ordering — new, old, random, key, rating, readinglog, want_to_read, currently_reading, already_read or title. Omit for relevance
• limit: Optional, 1–50, defaults to 10
• offset: Optional, 0–1000, defaults to 0

Example get_book_by_title input:

{
  "title": "The Hobbit",
  "limit": 1
}

Example get_book_by_title output:

{
  "num_found": 224,
  "offset": 0,
  "limit": 1,
  "results": [
    {
      "title": "The Hobbit",
      "authors": ["J.R.R. Tolkien"],
      "first_publish_year": 1937,
      "open_library_work_key": "/works/OL27482W",
      "edition_count": 481,
      "author_keys": ["OL26320A"],
      "best_edition": {
        "edition_key": "OL51709286M",
        "isbn_13": "9780395520215",
        "isbn_10": "0395520215"
      },
      "cover_url": "https://covers.openlibrary.org/b/id/14627509-M.jpg",
      "ratings_average": 4.29,
      "ebook_access": "borrowable"
    }
  ]
}

Example get_authors_by_name input:

{
  "name": "J. R. R. Tolkien",
  "limit": 2
}

Example get_authors_by_name output:

Each result's key can be passed to get_author_info for that author's full record. alternate_names is abridged here.

{
  "num_found": 2,
  "offset": 0,
  "limit": 2,
  "results": [
    {
      "key": "OL26320A",
      "name": "J.R.R. Tolkien",
      "alternate_names": ["John Ronald Reuel Tolkien", "Tolkien"],
      "birth_date": "3 January 1892",
      "top_work": "The Hobbit",
      "work_count": 355
    },
    {
      "key": "OL332676A",
      "name": "J. R. R. Tolkien Centenary Conference (1992 Keble College, Oxford)",
      "top_work": "Proceedings of the J.R.R. Tolkien Centenary Conference, 1992",
      "work_count": 2
    }
  ]
}

Example get_author_info input:

{
  "author_key": "OL26320A"
}

Example get_author_info output:

{
  "name": "J. R. R. Tolkien",
  "personal_name": "John Ronald Reuel Tolkien",
  "birth_date": "3 January 1892",
  "death_date": "2 September 1973",
  "bio": "John Ronald Reuel Tolkien (1892-1973) was a major scholar of the English language, specializing in Old and Middle English. He served as the Rawlinson and Bosworth Professor of Anglo-Saxon and later the Merton Professor of English Language and Literature at Oxford University.",
  "alternate_names": ["John Ronald Reuel Tolkien"],
  "photos": [6791763],
  "key": "/authors/OL26320A",
  "remote_ids": {
    "viaf": "95218067",
    "wikidata": "Q892"
  },
  "revision": 43,
  "last_modified": {
    "type": "/type/datetime",
    "value": "2023-02-12T05:50:22.881"
  }
}

Example get_author_photo input:

{
  "olid": "OL26320A"
}

Example get_author_photo output:

https://covers.openlibrary.org/a/olid/OL26320A-L.jpg

When Open Library has no photo for that author, the tool says so instead of returning a URL:

No author photo available for OLID OL99999999A.

Example get_book_cover input:

{
  "key": "ISBN",
  "value": "9780547928227",
  "size": "L"
}

Example get_book_cover output:

https://covers.openlibrary.org/b/isbn/9780547928227-L.jpg

As with author photos, a book with no cover produces a message rather than a URL:

No cover image available for OLID OL00000000M.

The get_book_cover tool accepts the following parameters:

• key: The type of identifier (one of: ISBN, OCLC, LCCN, OLID, or ID)
• value: The value of the identifier
• size: Optional cover size (S for small, M for medium, L for large, defaults to L)

Example get_book_by_id input:

{
  "idType": "isbn",
  "idValue": "9780547928227"
}

Example get_book_by_id output:

{
  "title": "The Hobbit",
  "authors": [
    "J. R. R. Tolkien"
  ],
  "publishers": [
    "Houghton Mifflin Harcourt"
  ],
  "publish_date": "October 21, 2012",
  "number_of_pages": 300,
  "isbn_13": [
    "9780547928227"
  ],
  "isbn_10": [
    "054792822X"
  ],
  "oclc": [
    "794607877"
  ],
  "olid": [
    "OL25380781M"
  ],
  "open_library_edition_key": "/books/OL25380781M",
  "open_library_work_key": "/works/OL45883W",
  "cover_url": "https://covers.openlibrary.org/b/id/8231496-M.jpg",
  "info_url": "https://openlibrary.org/books/OL25380781M/The_Hobbit",
  "preview_url": "https://archive.org/details/hobbit00tolkien"
}

The get_book_by_id tool accepts the following parameters:

• idType: The type of identifier (one of: isbn, lccn, oclc, olid)
• idValue: The value of the identifier

An example of this tool being used in Claude Desktop can be see here:

Docker

You can test this MCP server using Docker. To do this first run:

docker build -t mcp-open-library .
docker run -p 8080:8080 mcp-open-library

You can then test the server running within Docker via the inspector e.g.

npm run inspector http://localhost:8080

Development

Project Structure

• src/index.ts - The MCP server: builds the HTTP clients and drives both request handlers from the tool registry
• src/index.test.ts - Tests for the server wiring, including a snapshot of the published tool schemas
• src/tools/<tool-name>/ - One directory per tool, each containing index.ts (the handler, its Zod argument schema and its ToolDefinition), index.test.ts, and — for tools with a non-trivial API response — a types.ts describing that response shape
• src/tools/registry.ts - The TOOLS array, the single list of what the server exposes
• src/tools/types.ts - The ToolDefinition and ToolHandler contracts
• src/utils/ - Shared plumbing: http.ts (the API and covers Axios clients), errors.ts (argument parsing and error results), results.ts, schema.ts (Zod → JSON Schema), search.ts (the shared search projection and paging schemas), covers.ts
• scripts/ - Release automation (sync-server-json.mjs, promote-changelog.mjs, assert-release-consistency.mjs) and its tests

A tool's input contract is declared once, as a Zod schema. The JSON Schema that MCP clients see is generated from it by toInputSchema, so the two cannot drift. Field descriptions come from .describe() on the Zod schema. Note that .refine() constraints are dropped in translation — a cross-field rule has to be stated in the tool's description too, or clients will never learn about it.

Adding a tool means creating the directory and adding one entry to TOOLS in src/tools/registry.ts. src/index.test.ts derives its expectations from that array, so the only test change is an updated schema snapshot (npx vitest run -u).

Available Scripts

• npm run build - Build the TypeScript code
• npm run watch - Watch for changes and rebuild
• npm test - Run the test suite in watch mode
• npm run test:precommit - Run the test suite once and exit
• npm run lint / npm run lint:fix - Lint src and scripts with ESLint
• npm run format - Format code with Prettier
• npm run inspector - Run the MCP Inspector against the server

Running Tests

npm test starts Vitest in watch mode:

npm test

For a single pass — what the pre-commit hook and CI run — use:

npm run test:precommit

To run one file or one test case:

npx vitest run src/tools/get-book-by-id/index.test.ts
npx vitest run -t "should return book details when given a valid OLID"

Releasing

Releases are automated. Pushing a v* tag triggers publish-mcp.yml, which runs the checks, publishes the package to npm, registers the new version with the MCP Registry, and then creates a GitHub Release using that version's CHANGELOG.md section as the notes. Both npm and the registry authenticate over GitHub OIDC, so there are no publishing secrets to manage.

package.json's version is the single source of truth. npm version derives everything else from it via a version lifecycle hook, so a release is one command:

npm version patch   # or minor / major
git push --follow-tags

That single command bumps package.json, rewrites server.json to match, promotes the changelog's ## [Unreleased] heading to the new version and today's date, and commits the lot under one tag.

Two things to know before you run it:

• Write your changelog entries first. They go under a ## [Unreleased] heading in CHANGELOG.md as you merge work. npm version fails if that heading is missing, rather than releasing something undocumented. If it does fail, undo the partial bump with git restore --source=HEAD --staged --worktree package.json package-lock.json server.json.
• The working tree must be clean, and the pre-commit hook (lint + full test suite) runs inside npm version.

CI re-asserts that the tag, package.json, server.json and CHANGELOG.md all agree before anything is published — see scripts/assert-release-consistency.mjs. The same check runs on pull requests that touch those files.

npm version is not a retry

Once it prints the new tag, the commit and tag exist and the release is done locally — the next step is git push --follow-tags, not running npm version again. A second run attempts the next version, and will fail on the missing ## [Unreleased] heading (which the first run consumed). That failure is safe by design, but it leaves package.json, package-lock.json and server.json bumped and uncommitted. Undo with:

git restore --source=HEAD --staged --worktree package.json package-lock.json server.json

If the publish workflow fails

Re-running the job from the Actions tab only helps for a transient failure. GitHub runs the workflow as it existed at the tagged commit, so a bug in the workflow itself or in server.json cannot be fixed by a re-run — the fix has to be in the commit the tag points at.

Nothing is published until the workflow reaches its npm step, so if it failed before then, the version is still free and you can move the tag:

# fix the problem on main and commit it first
VERSION="$(node -p "require('./package.json').version")"

git push origin ":v${VERSION}"            # delete the remote tag e.g. git push origin :v1.0.3
git tag -d "v${VERSION}"                  # delete it locally
git tag -a "v${VERSION}" -m "${VERSION}"  # re-tag at the fixed commit
git push origin "v${VERSION}"

The fix commit must leave package.json on that same version, or the consistency check will reject the tag. If npm did already publish, do not reuse the version — that release is immutable. Bump to the next patch instead; the guarded npm step means a re-run skips what already succeeded.

Contributing

Contributions are welcome! Please feel free to submit a pull request.

Acknowledgments

• Open Library API
• Model Context Protocol