Epstein-Files-Plugin

flowchart LR
    subgraph Client["Any MCP client"]
        A["Claude Code / Claude Desktop\nCursor · Windsurf · custom agent"]
    end

    subgraph Server["epstein-files-plugin — local stdio, no auth"]
        direction TB
        PB["Purpose-built tools\nefta_search · efta_filter_email\nefta_lookup · efta_known_docs"]
        GEN["Generic tools\nefta_list_databases · efta_list_tables\nefta_describe_table · efta_query_table\nefta_run_sql"]
        LOCAL["Local-only\nefta_get_url — pure computation,\nno network call"]
    end

    subgraph Upstream["Third-party mirror"]
        DS[("epstein-data.com\nDatasette JSON API\n20 databases")]
    end

    DOJ[("justice.gov/epstein\nofficial PDF source")]

    A -- "MCP / stdio\nJSON-RPC" --> Server
    PB --> DS
    GEN --> DS
    LOCAL -.->|"constructs URL,\nno request sent"| DOJ
    DS -.->|"indexes"| DOJ

    style Server fill:#1a2332,stroke:#4b8bbe,color:#e8edf4
    style Upstream fill:#0f1720,stroke:#2dd4a7,color:#e8edf4
    style DOJ fill:#0f1720,stroke:#94a3b8,color:#e8edf4

/plugin marketplace add Zereo0317/Epstein-Files-Plugin
/plugin install epstein-files-plugin@epstein-files-plugin

git clone https://github.com/Zereo0317/Epstein-Files-Plugin.git
cd Epstein-Files-Plugin
python -m pip install --upgrade pip
pip install -r requirements.txt   # requests, mcp (pinned <2.0 — see below)

python src/efta_researcher.py --list
python src/efta_researcher.py --search "trilateral commission"
python src/efta_researcher.py --sender epstein --recipient schank --date 2009-10-23
python src/efta_researcher.py --efta EFTA00741068
python src/efta_researcher.py --databases
python src/efta_researcher.py --tables image_analysis
python src/efta_researcher.py --sql "select dataset, count(*) as n from doc_search group by dataset" --database full_text_corpus

python src/mcp_server.py

{
  "mcpServers": {
    "epstein-files-plugin": {
      "command": "python",
      "args": ["/absolute/path/to/epstein-files-plugin/src/mcp_server.py"]
    }
  }
}

{
  "mcpServers": {
    "epstein-files-plugin": {
      "command": "python",
      "args": ["${CLAUDE_PLUGIN_ROOT}/src/mcp_server.py"],
      "env": { "PYTHONUNBUFFERED": "1" }
    }
  }
}

User request:  "What's the DOJ URL for EFTA00741068, and which DataSet is it in?"

Response:
  EFTA: EFTA00741068
  DataSet: DS9
  URL: https://www.justice.gov/epstein/files/DataSet%209/EFTA00741068.pdf

Under the hood:
  efta_get_url("EFTA00741068") parses the Bates number, resolves it against the verified
  12-DataSet boundary table, and constructs the official DOJ URL — no network call, no guessing.

User request:  "Find the Epstein -> Roger Schank email from October 23, 2009."

Response:
  Found 3 email(s):
    EFTA00741068  DS9   2009-10-23  12:01:12   <- primary
    EFTA00885615  DS9   2009-10-23  12:01:12   (OCR duplicate, reads "grnail.com")
    EFTA01821140 DS10   2009-10-23  12:01:12   (third copy, later processing batch)

Under the hood:
  efta_filter_email(sender="epstein", recipient="schank", date_exact="2009-10-23") queries the
  epstein-data.com Datasette API (DOJ's own /multimedia-search is blocked for headless clients)
  and returns every metadata match, so duplicate copies can be cross-referenced by timestamp.

User request:  "Any photos in the release that show a passport?"

Response:
  efta_list_tables("image_analysis") -> "images" table, 92,249 rows, column "analysis_text"
  efta_query_table("image_analysis", "images", filters={"analysis_text__contains": "passport"})
  -> matching rows with efta_number, source_pdf, and the analysis text itself

Under the hood:
  image_analysis isn't reachable through efta_search (that only covers full_text_corpus).
  efta_query_table works against any of the 20 databases using the same filter-suffix syntax,
  so no dedicated "image search" tool was needed.

User request:  "Break down the document count by DataSet."

Response:
  efta_run_sql("full_text_corpus",
    "select dataset, count(*) as n from doc_search group by dataset order by dataset")
  -> 1:650, 2:150, 3:57, 4:143, 5:82, 6:13, 7:17, 8:10479, 9:480658, 10:496404,
     11:331597, 12:12339, 98:6, 99:23210   (live counts, 2026-08-24)

Note: two values (98, 99) fall outside the documented 1-12 DataSet scheme — small catch-all
buckets in the source data itself. efta_get_url()/get_dataset() only resolve DataSets 1-12.

.claude-plugin/plugin.json   Plugin manifest (Claude Code convenience only)
.mcp.json                    Local stdio MCP server config (client-agnostic)
src/efta_core.py             DataSet boundary table, EFTA -> URL, known-document registry
src/epstein_datasette.py     epstein-data.com Datasette API client (purpose-built + generic layer)
src/doj_auth.py              justice.gov public anti-bot challenge helper + verification
src/efta_researcher.py       Standalone CLI
src/mcp_server.py            FastMCP server exposing the 11 tools above
skills/efta-research/        Claude Code skill: research methodology (optional convenience)
skills/doj-auth/             Claude Code skill: justice.gov access details (optional convenience)