mcp-books is an MCP (Model Context Protocol) server for querying three large digital archives at once: the Internet Archive, the Library of Congress, and the Bibliothèque nationale de France (via data.bnf.fr). It requires no API key or account, and can search text within scanned documents, search catalogs, and read individual records in a unified format.
Main capabilities
- Cross-archive search: query all three archives with a single request, with each archive responding independently.
- Scanned text search: search text within scanned documents processed by optical character recognition (OCR), returning hits that include id, source, title, creator, year, page number, excerpt, and excerpt type (excerpt_kind).
- Catalog search: search the catalogs of each archive. Different archives classify materials differently; the Internet Archive searches all types when no type is specified; the Library of Congress has a separate path for each type; BnF searches works.
- Record reading: read a single record in a unified format, regardless of which archive it comes from. Returns id, source, the archive's own identifier, title, creator, date, year, and the meaning of the year (year_means), attribution information (attribution), etc.
- Independent reporting: the results of each archive are reported separately; failures, unqueried archives, and empty results are identified separately, and counts are not added across archives.
How to use
- One-click install: supports Claude Desktop, Cursor, and any client that uses the standard configuration format. Requires Node 24 or later, with no environment variables to set.
- Bundle install: download mcp-books-2.0.1.mcpb from the latest release and open it; clients that support MCP bundles will install it automatically, with no npm or configuration file needed.
- Configuration: all variables are optional and can be set in the env block of the client configuration. Values outside the range fall back to the default, and the reason is written to stderr.
- Network requirements: the container needs outbound HTTPS access to archive.org, openlibrary.org, www.loc.gov, and data.bnf.fr, with no volumes, ports, or credentials required.
Suitable scenarios
- Find books that mention a specific place or event, for example "Which books mention the Beaumont lighthouse?" or "Find materials about the 1906 San Francisco earthquake."
- Read records and learn who holds the original materials.
- Query BnF for materials about an author.
- Search for photos rather than books.
Limitations and notes
- Responses take several seconds, because the three archives are each queried at their own pace.
- Scanned text comes from OCR, and excerpts may contain recognition errors; they are provided as-is, and when citing them you should note that they are scanned text and link to the record.
- Each archive has its own request pace, and the Library of Congress is the slowest; when querying all three archives at once, each archive receives only one request.
- This project is unofficial and is not affiliated with the archives it reads.
- The server does not collect user information, does not send data to the authors, and keeps answers only in memory, without writing to disk.
- Tests use generated fixed data and do not make network requests; the live test suite runs every night.
- The license is MIT, and the records belong to their respective archives and their holders.
