Epstein MCP server

epstein

Full-text search over the documents the House Oversight Committee has published for its Jeffrey Epstein investigation — transcribed interviews, subpoenas and letters — plus (phase 2) a structured search over Epstein's own pilot flight logbook. Fleet #1480, phases 1-2.

Part of Pipeworx — an MCP gateway connecting AI agents to 1558+ live data sources.

Tools

Tools are prefixed because search_documents is already taken by federal-register and riksdagen-se; a colliding tool ships unreachable while the error advertises a namespaced name that does not resolve.

Phase 2: the flight logbook

epsteinflightlogsearch reads Government Exhibit 662-RR — Epstein's own pilot flight logbook (1991-2005), admitted into evidence and unsealed in USA v. Ghislaine Maxwell*, 20-cr-330 (S.D.N.Y.), hosted keylessly at a stable DocumentCloud asset URL. Phase 1's original release note said no flight logs existed anywhere in the House Oversight corpus — true, but that was never the only public source. This exhibit is a fixed, closed 118-page historical document, not a live feed, so ingest is a one-time transcription rather than a per-request proxy call.

397 of an estimated 3,000+ logged legs are ingested — 23 of the 118 pages (fleet #1880 added pages 4-13 to the original 10-page seed, which was chosen to span the full 1991-2005 range; fleet #1881 added pages 14-16). Machine OCR of this handwritten logbook is unusable (the exhibit's own bundled OCR text layer is illegible line noise, verified before building this); every ingested row is a direct visual transcription of the page image. epsteinflightlog_search with no filters returns the exact pages covered so an absence never reads as "did not fly" — it may simply mean the page is not yet transcribed. Completing the remaining 95 pages is follow-on work, not a blocker to shipping this slice.

Confidence convention on the transcription itself (fleet #1880): passengers_raw is verbatim per row — the same person's name is NOT normalized to one spelling across rows even where it clearly varies ("Krestena" / "Kristena" / "Christina" all occur for what is plausibly one person). A token read with real but not total confidence carries a trailing [?] inside the word it qualifies ("Arthur Hallering[?]"); a token that could not be read at all is [illegible]. Neither is a guess standing in for a fact. A pilot's own struck-through self-correction is omitted rather than marked [redacted] — that marker is reserved for a black-out on the original exhibit page.

Design decision C is held exactly: passenger entries are never resolved. "JE", "GM", "SK" and similar initials are stored and returned exactly as the pilot wrote them, never expanded to a full identity — even where that identity is publicly reported elsewhere. A black-out on the original page is stored as the literal marker [redacted]; a word the transcriber could not read with confidence is stored as [illegible]. Neither is a guess standing in for a fact.

Every row carries its own provenance: exhibit title, the page number as printed on the page itself, the DocumentCloud source_url, and the sha256 of the source PDF, so a caller never has to trust that two rows came from the same file.

What this is, and what it is not

This is one committee's published PDFs. It is not "the Epstein files", and every response says so rather than letting the name imply completeness.

Shortened. The full README is on GitHub.