Georgia Archives MCP server

@pipeworx/georgia-archives

The three public catalogues of the National Archives of Georgia — the scientific reference library (Koha), the national fonds register, and the Georgian national film catalogue.

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

Tools

by keyword / title / author / subject. Returns a true total (not a page count) plus 20 records, each with its permanent biblionumber and catalogue URL.

• archiveslibrarysearch(query, index?, offset?) — search the reference library

field/subfield structure plus a flattened summary (title, authors, imprint, ISBN, languages, subjects).

• archiveslibraryrecord(biblionumber) — one record as parsed MARC: the full

pages of 25. Each row carries the holding regional archive, the fond number, the fond name, the historical names it held with their date ranges, and the inclusive dates of its administrative and personnel records.

• archivesfondsbrowse(page?) — walk the fonds register: ~14,650 fonds over 586

newsreels, documentaries and features, with directors and camera operators as {id, name} objects carrying the catalogue's own person ids.

• archivesfilmsearch(name?, page?, locale?) — the film catalogue's 652

Auth

Keyless. Every endpoint answers without a credential.

Data sources

— Koha OpenSearch RSS. is an honest total and &offset=N is honoured, so pagination has a real cursor.

/api/v1/ itself is a keyless Swagger 2.0 spec, 129 paths, 13 of them public.

• — Koha public REST.

catalogue, an Inertia/Laravel app that returns JSON when asked.

• — the fonds register.
• — the film

Things the next person would otherwise rediscover the hard way

Koha's biblio route rejects Accept: application/json with a 400. It requires a MARC media type — application/marc-in-json or application/marcxml+xml. Test it with default headers and you will file a working endpoint as dead.

The fonds register's trailing slash is load-bearing. /fonds?page=2 301s to /fonds/ and drops the query, so you get page 1 back with a 200. Its search is POST-only behind a token CSRF field, and the same parameter names sent over GET are silently ignored* — page 1 again, which reads as a clean zero-result. There is no JSON surface here at all (main.js greps out to jQuery internals, zero endpoints). Browse-and-index is the only honest shape; do not build a search proxy on it.

The film catalogue ignores unknown parameters instead of rejecting them. search=, keyword= and title= all return the unfiltered 652 with a 200. Only name filters. That is why this pack accepts exactly one filter and nothing else — an accepted-but-ignored argument would hand the caller the whole catalogue dressed as a search result.

X-Inertia-Version is a deploy fingerprint, not a constant. Hard-code it and you get a 409 the next time they deploy. currentInertiaVersion() reads it out of the page's data-page attribute on every call, which costs one extra request and cannot go stale.

OAI-PMH is not enabled on the Koha instance (/cgi-bin/koha/oai.pl?verb=Identify → 404), so there is no bulk path; the REST + OpenSearch pair is the whole surface.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "georgia-archives": {
      "url": "https://gateway.pipeworx.io/georgia-archives/mcp"
    }
  }
}

What this endpoint actually serves

Shortened. The full README is on GitHub.