Epstein

What Epstein does

Epstein provides MCP tools for examining documents published by the U.S. House Committee on Oversight and Government Reform for its Jeffrey Epstein investigation. The corpus contains 31 documents totaling 505 pages and roughly 800,000 characters, including transcribed interviews, subpoenas, and letters. Search results identify the document, page number, and the committee URL so a reader can verify each match.

A separate tool searches a structured transcription of Government Exhibit 662-RR, Epstein's pilot flight logbook from 1991–2005. The available transcription covers 397 logged legs across 23 of the exhibit's 118 pages. It preserves passenger entries as written, including initials, uncertain readings, and literal redaction or illegibility markers.

How it works

The server is available through the Pipeworx MCP gateway at https://gateway.pipeworx.io/epstein/mcp. Add that URL as an MCP server in a compatible client, then invoke the tools through the client’s normal MCP interface.

The committee-document corpus is harvested from Oversight release pages and the PDFs linked from those pages. A weekly refresh workflow re-harvests the source as the committee publishes additional material. The flight-log exhibit is a fixed historical document hosted on DocumentCloud and was transcribed once rather than fetched for every request.

Flight-log rows include the exhibit title, printed page number, source URL, and SHA-256 hash of the source PDF. An unfiltered flight-log search returns the pages currently covered by the transcription, helping distinguish an uncovered page from evidence that a flight did not occur.

Setup and configuration

Configure the hosted endpoint in an MCP client with an entry such as:

{
  "mcpServers": {
    "epstein": {
      "url": "https://gateway.pipeworx.io/epstein/mcp"
    }
  }
}

The quick-start configuration uses no credentials or environment variables. The endpoint can be added to Claude Desktop, Cursor, or Windsurf.

Tools and capabilities

• epstein_search_documents searches the text corpus and returns page-level citations.
• epstein_list_documents describes the documents in the corpus and indicates which are searchable.
• epstein_get_document returns the full text of a searchable document one page at a time.
• epstein_flight_log_search filters the ingested flight-log transcription by tail number, route, or date range.

The tool names use an epstein_ prefix to avoid collisions with similarly named tools exposed by other MCP servers.

Limitations and notes

This collection is limited to documents published by the House Oversight Committee and does not represent every document commonly described as the “Epstein files.” Court filings and dockets are handled separately by court-listener, and the estate-document tranche linked from a later committee release is not available from this source.

Four of the 31 committee documents are scans without text layers. They appear in the document listing as non-searchable, and document retrieval refuses those files while providing a link to the original. Search coverage therefore should be checked with epstein_list_documents before treating a missing match as conclusive.

The flight-log search covers only 23 of 118 pages, so an absent result may reflect an untranscribed page. Passenger names and initials are not normalized or resolved to identities. The transcription retains uncertainty markers such as [?], [illegible], and [redacted]; it should be read as a source transcription, not an identity-resolution or relationship-analysis system.

The server intentionally provides citations and document content rather than inferred associations, relationship scores, entity pages, or claims about people named in the records. It is an independent integration and is not affiliated with, endorsed by, or published by the upstream sources.