Nplg Catalog MCP server

@pipeworx/nplg-catalog

The union catalogue and the digital library of the National Parliamentary Library of Georgia — Georgian bibliographic records with their MARC intact, plus a small full-text digital collection.

Part of Pipeworx — an MCP gateway connecting AI agents to 1576+ live data sources.

Tools

keyword / title / author / subject. Returns a true total (not a page count) plus 50 brief citations, each with its permanent Sierra bib_id, title (Georgian with the parallel Latin title split out where the record carries one), author, imprint, material type and a stable catalogue URL. scope 1-11 narrows by material — books, periodicals, maps, dissertations, archival documents, newspaper and journal articles.

• nplg_search(query, index?, scope?, offset?) — search the union catalogue by

and every data field with its indicators and subfields kept separate, plus a flattened summary (title, authors, imprint, languages, classification, subjects, host item). The 041 language codes and the parallel Georgian/English 245 survive as structure rather than as one string.

• nplgrecordmarc(bib_id) — one record as parsed MARC: leader, control fields,

print archive, dissertations and posters, each hit with authors, date of issue, a full source citation, subject headings in Georgian/Russian/English, a Dewey subject and a stable item URL.

• nplgdlibrarysearch(query, collection?, page?) — the digital library: scanned

Auth

Keyless. Every request answers without a credential.

A Sierra REST API v6.1.0 is live at catalog.nplg.gov.ge/iii/sierra-api/v6/ and 401s without a library-issued client key/secret (OAuth2 client credentials, exchanged at /v6/token). Asking NPLG for one would turn every HTML parse in this pack into a supported JSON API and change nothing else about its shape.

Data sources

union catalogue, Innovative Interfaces Sierra behind the classic WebPAC Pro front end (Server: III 100).

• — the

— III's native offset cursor. — per-record MARC, | as the subfield delimiter.

• https://catalog.nplg.gov.ge/search~S*eng?/X…/%2C%2C%2CB/browse

library. co is the collection (0 all, 1 print archive, 2 dissertations, 3 posters), pg the 1-based page of 10.

• — the digital

Things the next person would otherwise rediscover the hard way

The phrase indexes do not return records. /searcheng/t, /a and /d answer 200 with an alphabetical list of headings* and zero brief citations — a title search that looks like it returned nothing. Records come from the keyword index with III's own field qualifier inside the query: t:(tbilisi), a:(abashidze), d:(tbilisi). That is what index: does here.

The offset cursor carries the result total in the URL, and lying to it still returns rows. The 51,2896,2896,B triple is first,last,total. Pass a total that is not the search's and you still get a 200 with records, under a header reporting a total that belongs to nothing — a page that is right about the rows and wrong about the size of what it is paging through. So an offset request reads the real total from the first page before asking for the offset; it costs one extra request and is the only way to page honestly.

The ~S in the cursor path must match searchscope=. They are the same number in two places and the path one is what actually scopes the browse.

Georgian arrives as numeric character references. Most of the markup is მ rather than UTF-8, and MARC punctuation comes through as &#59; / ". An entity decoder that only knows & and friends returns mojibake that still parses cleanly.

Shortened. The full README is on GitHub.