184,000+ primary sources — peer-reviewed research from 50 top universities, digitized history from the New York Public Library, semiconductor manufacturing training, and arXiv preprints — searchable by any MCP-compatible agent. Free. Fully attributed. No authentication.
Seven collections, one search. Every item carries a source_url back to the originating institution — full primary source, fully attributed, ready for citation.
| Collection | Items | Topic sets |
|---|---|---|
| OpenAlex peer-reviewed CS & engineering papers | 165,876 | 60 sets — Machine Learning, AI, NLP, Computer Vision, Cybersecurity, Algorithms & more |
| NYPL digitized historical records | 18,541 | 17 sets |
| American-history curated sets (via DPLA) | — | 40 sets — Civil Rights, WWII, Women's Suffrage, Great Depression, Space Race & more |
| Semiconductor manufacturing training | 224 | 7 curriculum sections |
| arXiv preprints | 192 | 12 sets |
| Library of Congress (live search) | — | 10 research sets |
| One Bite pizza reviews (demo dataset) | 1,143 | 1 set — try search_pizza_reviews |
MIT's 6.780 curriculum taught interactively by AI using the training collection above. Lectures, problem sets, and quizzes — no classroom, no tuition.
The web is quietly closing to AI. Bot-blocking, paywalls, and anti-crawling defaults mean a growing share of human knowledge returns a 403 when an AI agent asks for it — even when the content itself is public, open-access, or in the public domain. Every site that goes dark to agents makes AI answers a little worse for everyone.
This library is the opposite bet: primary sources — peer-reviewed research, digitized history,
open training materials — deliberately kept reachable, structured, and attributed for any AI
agent that asks. Not scraped content laundered from someone else's work, but open collections
served with a source_url
on every item, so the originating institution gets the credit and the reader gets the primary source.
Wikipedia became the open reference layer of the human web — free, cited, and built to be read by anyone. The AI-native web needs the same thing: a free, cited reference layer built to be queried by any agent. That's what this library is for.
Full-text BM25-ranked search across titles, abstracts, and document text — all 184,000+ items at once, or filtered by collection with source="micron-training".
Pull the full record for any item by ID — title, description, item type, institution, and the source_url for citation.
Enumerate all 140 curated topic sets across every collection — the map of what the library holds, filterable by source.
Query the New York Public Library and Library of Congress digital collections live — reaching beyond the indexed corpus to millions of additional records.
Once connected, paste any of these into a new conversation.
Free, no catch. No API key, no authentication, no rate limits on reads. The library is the open content layer of the PUBLICMCP standard — it exists so any agent can see what a public MCP server is capable of. If you find it useful, look at the business directory built on the same infrastructure.
Any agent that speaks the Model Context Protocol: Claude (including free claude.ai accounts via Settings → Connectors), the Gemini API, Ollama, and custom agent frameworks. ChatGPT integrations can use the REST endpoint at /tools/{tool_name}.
Open scholarly and cultural sources: OpenAlex (peer-reviewed papers from 50 top universities), the New York Public Library's digital collections, the Digital Public Library of America, the Library of Congress, arXiv, and curated open training materials from OSHA, SEMI, and Micron Technology. Every item links back to its originating institution.
Yes — PUBLICMCP hosts MCP servers free for public libraries and educational nonprofits, on the same infrastructure that serves this library. Contact us through publicmcp.org.
Every set has its own page with a sample of the collection and the MCP query to retrieve all of it. Agents: markdown index at /library/index.md.
