ShelfLayer gives your model the right book, the right passage, and a citation it can carry back into the conversation.
We commissioned a series of portraits of our regular patrons. The faces are theirs. The chairs, the bookplates, and the right-cuffed sleeves are ours.
The Scholar · in the reading chair
Agents that draft literature reviews, essays, and survey notes can search public-domain books first, then quote with a source trail attached.
The Practitioner · between consultations
Coding agents, chat clients, and internal copilots that need a library tool instead of another unstructured web scrape.
The Investigator · checking a source
Teams that want a custom MCP pointed at internal docs, a company wiki, country law, technical manuals, or another private corpus.
For two thousand years, humanity has been hiding its best thinking inside books. Then, in a span of twenty years, we taught machines to ignore them — to graze instead on a paddock of search-optimised paragraphs, ad-laced summaries, and confidently-wrong forum posts.
We disagree with that arrangement. The deepest answer to most questions worth asking has already been written, edited, peer-reviewed, and shelved. What's missing is the librarian — one who can listen to an agent's task, walk the stacks, and return with the right chapter open.
ShelfLayer is that librarian, exposed as an MCP server.
A thin gruel of listicles, content farms, AI-generated spam, paywalled previews, and reddit threads from 2014 — all weighted by who paid most for the click.
Public-domain books with shelves, tables of contents, sections, notes, and passages. The beta starts with Project Gutenberg and expands from there.
From your agent's vague intention to a citable paragraph, in less time than it takes to walk to a shelf.
Tell us the subject as a working librarian would understand it — "stoic ethics," "political economy," "early psychology." We resolve it to shelves and book metadata.
We search the indexed passages with book-aware filters, then traverse the strongest matches by shelf, book, and section — never just titles, never just tables of contents.
Your agent receives the passage, the surrounding section when needed, and a structural citation. The kind of source trail that survives inspection.
Connect your MCP host to ShelfLayer and call search_passages, list_books, inspect_book, read_section, and get_citation.
# your agent needs source material on thrift and interest. # it asks the shelf instead of scraping the web. { "name": "search_passages", "arguments": { "query": "\"compound interest\" OR thrift economy", "shelf": "economics-and-finance", "limit": 5 } } // → passages with book_id, section_id, passage_id, // plus a citation helper for user-facing answers.
Free during public beta. Connect with Google from your MCP client.