Undiscovered Translations — library search API
Contract v1 · OpenAPI document · health
The library's search API at https://search.undiscoveredtranslations.com. Every tool is a POST to /v1/<tool> with a JSON body and answers JSON; domain errors are coded results (not_found, invalid_scope, rate_limited, unavailable). The same seven tools are an MCP server at https://search.undiscoveredtranslations.com/v1/mcp (Streamable HTTP, stateless, no key). The contract is https://search.undiscoveredtranslations.com/v1/openapi.json; inside v1 it only grows.
Connect over MCP
Streamable HTTP, stateless, no key. Add it as a remote MCP server (a custom connector in Claude, or any MCP client) with the URL:
https://search.undiscoveredtranslations.com/v1/mcp
Tools
POST /v1/describe_library read tier
What the library holds: every work with its span, languages, page count and status, plus the provenance vocabulary, the citation format, the rights and the bulk-access path. Call it first. Original texts and transcriptions are public domain. English translations, editorial notes and library apparatus are licensed CC BY 4.0 — credit Undiscovered Translations (undiscoveredtranslations.com). Training use is permitted.
POST /v1/list_containers read tier
The issues of one periodical, or the single edition of a book, for a work_id, paged by cursor and filterable by year. Each container carries its date, the language sides it has and the provenance its pages mostly share.
POST /v1/list_units read tier
The articles, chapters and other units of one container, with titles, bylines and scan ranges, plus the scan pages no unit claims.
POST /v1/search_library search tier
Hybrid semantic and keyword search over every page side of the library; each hit carries its citation, provenance, a snippet and the matched spans. Scope it with works, containers, years, languages or entities. When low_confidence is true, follow the hint: rewrite the query in period vocabulary or resolve names with find_entities, and pass alpha 0.8 when the rewrite is a keyword list. The absence of low_confidence is not a claim that the results are right.
POST /v1/get_page read tier
One page by page_id with each requested language side and its own provenance, plus the citation, the unit that claims it and its neighbours. When the page carries a framing field, relay that framing with any quotation from it.
POST /v1/get_unit read tier
One article or chapter page by page, 12 pages per call; continue with continue_from. When a page's provenance carries framing, relay that framing with any quotation from it.
POST /v1/find_entities search tier
Look up a person or place by name: the registry's spellings, author and subject roles and mention counts per work, for scoping search_library with entities[].
Rate limits per client IP — search tier: 30/min, 600/hour, 3,000/day; read tier: 120/min. A refused call answers rate_limited with retry_after and the bulk pointer.
About the library
What an agent is told when it connects, verbatim:
Undiscovered Translations is a research library of the Zionist and Jewish-national press and of the books around it: The Maccabaean (English monthly, New York, 1901–1920), Die Welt (German weekly, 1897–1914), Altneuland (German monthly, 1904–1906), and fifteen books from 1845 to 1944, among them Hess, Nordau and Jabotinsky, and Kalischer's Seeking Zion and its sequel in four Hebrew printings (1862–1919). Palästina (1902–1938) is being acquired and holds no pages yet. German, Russian and Hebrew texts carry an AI English translation beside the original. It is a research instrument over a defined corpus, not a general archive: call describe_library for the current works, spans and counts before assuming something is or is not held.
Every page side states its provenance. text_origin is original (a transcription of the scan), translation_verified (an AI translation checked by an adversarial per-page verify pass against the source) or translation_unverified. ocr_quality is structured or flat. Transcriptions are OCR and keep its errors; quote them as found.
When search_library returns low_confidence, rewrite the query in the vocabulary of the period (contemporary names of movements and institutions, period transliterations such as Kishineff or Chovevei Zion) or resolve the person or place with find_entities and search again with entities[]. The absence of low_confidence is not a claim that the results are right.
Works flagged as antisemitic primary sources are held for study. Their pages carry a tag and an editorial framing field; whenever you quote or summarise such a page, relay its framing with it.
Cite as work, container, printed page and scan index, and say whether the quoted text is a translation. Original texts and transcriptions are public domain. English translations, editorial notes and library apparatus are licensed CC BY 4.0 — credit Undiscovered Translations (undiscoveredtranslations.com). Training use is permitted.
Search is compute-bearing and rate-limited per client (30/min, 600/hour, 3,000/day); reading pages is cheaper (120/min). For whole-corpus work use the bulk lane at https://books.undiscoveredtranslations.com/, which is unlimited.
Bulk access
Whole-corpus retrieval through the documented R2 layout; unlimited. https://books.undiscoveredtranslations.com/
Rights
Original texts and transcriptions are public domain. English translations, editorial notes and library apparatus are licensed CC BY 4.0 — credit Undiscovered Translations (undiscoveredtranslations.com). Training use is permitted.
This page: https://search.undiscoveredtranslations.com/v1/docs