ShelfLayer is an MCP server that gives AI agents a library instead of a search engine. When your agent needs to know what Adam Smith or Marcus Aurelius actually wrote, it can pull the real passage from the book rather than paraphrasing whatever turned up on the web.
What’s on the shelf
- Almost 28,000 public domain books from Project Gutenberg
- More than 20 million indexed passages
- Search by subject, book, author, section, or a partial phrase you half remember
How agents use it
You connect it to any MCP client (Claude, Cursor and so on) and sign in with Google. It exposes eight read-only tools, including search_passages, list_books, inspect_book and read_section. The agent gets back the exact passage with a citation that points to where it lives in the book, so there’s no “the author said something like this” and no made-up quotes.
Their tagline is “Stand on the shoulders, not the search results,” which is pretty good.
Pricing
Free while it’s in public beta.
Why I saved it: hallucinated citations are one of the most annoying things about AI research tools. Grounding an agent in actual source text is a simple, smart fix, at least for anything old enough to be public domain.