Icij Offshore Leaks MCP Server

@pipeworx/icij-offshore-leaks

Search the ICIJ Offshore Leaks Database — offshore entities, officers (directors/shareholders/beneficiaries) and intermediaries (registered agents, law firms, banks) named across five ICIJ investigations (Offshore Leaks 2013, Panama Papers 2016, Bahamas Leaks 2016, Paradise Papers 2017, Pandora Papers 2021) — by name, or fetch one node's full profile plus every linked officer/intermediary/entity/address and which leak each link came from.

Part of Pipeworx — an MCP gateway connecting AI agents to 1807+ live data sources. This is an independent, unofficial integration — not affiliated with, endorsed by, or published by the upstream provider.

Tools

• offshore_search(name, jurisdiction?, node_type?, limit?) — name search (prefix-aware FTS5) across entities, officers and intermediaries, with optional jurisdiction and node-type filters. Returns each match's node_id, type, jurisdiction, status and which leak it came from.
• offshore_entity(node_id, limit?) — one node's full profile (name, jurisdiction, status, incorporation/struck-off dates, registered agent) plus every linked relationship (role, direction, the other node's name and type, and the leak it comes from). Works for any node type.

Auth

Keyless.

Licence and caveat — every response carries both, verbatim

The ICIJ Offshore Leaks Database is licensed under the Open Database License (ODbL) (contents under Creative Commons Attribution-ShareAlike). ICIJ: "Always cite the International Consortium of Investigative Journalists when using this data."

ICIJ: "The inclusion of a person or entity in the ICIJ Offshore Leaks Database is not intended to suggest or imply that they have engaged in illegal or improper conduct." Many people and entities share the same or similar names.

This pack serves only what ICIJ itself already publishes — names, roles, jurisdictions and links. No enrichment with any other personal-data source.

Data source

• International Consortium of Investigative Journalists (ICIJ), Offshore Leaks Database (offshoreleaks.icij.org) — ICIJ's bulk CSV export of the graph behind its public database: ~814K entities, ~771K officers, ~27K intermediaries, ~402K addresses, ~3K other nodes, and ~3.34M relationship edges between them, across five point-in-time investigations.

Store, not a bundled import, not a live query. ICIJ publishes this as a bulk CSV download under ODbL, not a callable API, so this is a BUILD-THE-COPY pack (CLAUDE.md "What we will source"). The ~2.0M nodes and ~3.3M edges live in a D1 database the gateway hands this pack as _d1 ("d1Binding": "ICIJ_OFFSHORE_LEAKS_DB" in pack-manifest.json; schema + queries in src/store.ts) — at this size, a bundled import would have blown the gateway Worker's bundle ceiling outright (28 MiB raw / 6.5 MiB gzip; this dataset alone is ~1 GB of SQL).

A known data quirk, handled in the loader: ICIJ's export reuses the same node_id for ~1,139 people across nodes-officers.csv AND nodes-intermediaries.csv (identical name/country/source — e.g. node_id 51122, "Peng, Wan-Hsiung", filed as both). ICIJ's own site (offshoreleaks.icij.org/nodes/<id>) resolves those ids to "Intermediary", so the loader keeps the intermediary row and drops the officer duplicate for any id that collides (scripts/icij-offshore-leaks/fetch-icij-data.py).

Search is FTS5, not LIKE — unlike smaller D1-store packs in this catalog, ~1.6M searchable names is too many for a LIKE '%term%' scan to answer reliably inside D1's per-query limits. offshore_name_fts is a standalone FTS5 table (not content=/content_rowid=), decoupled from offshore_node's rowid and joined back by the node_id TEXT column — chosen specifically to avoid the two known D1 FTS5 traps: a chunked base-table import never populates FTS5 postings (so the loader rebuilds the index with its own batched statements, server-side, AFTER the base load is verified complete), and an external-content FTS naming a column the base table lacks errors live on D1 while staying silent under a local sqlite smoke test.

Refresh:

python3 scripts/icij-offshore-leaks/fetch-icij-data.py --out /tmp/icij-ndjson
node scripts/icij-offshore-leaks/load-data.mjs --ndjson /tmp/icij-ndjson --db pipeworx-icij-offshore-leaks

ICIJ updates the bulk export on its own schedule (five point-in-time investigations since 2013, most recently Pandora Papers in 2021, with incremental corrections since). data_as_of on every response is the export's own generation date as of the last load. This writes to production D1 — only run it with that access; see "Production setup" below for the one-time database creation this depends on.

Production setup (one-time, not done by this pack's author — D1 writes are Bruce's call)

wrangler d1 create pipeworx-icij-offshore-leaks --location wnam
# then add the [[d1_databases]] block it prints to workers/gateway/wrangler.toml:
#   binding = "ICIJ_OFFSHORE_LEAKS_DB"
#   database_name = "pipeworx-icij-offshore-leaks"
#   database_id = "<from the create output>"

python3 scripts/icij-offshore-leaks/fetch-icij-data.py --out /tmp/icij-ndjson
node scripts/icij-offshore-leaks/load-data.mjs --ndjson /tmp/icij-ndjson --db pipeworx-icij-offshore-leaks

Verified locally against the real 2026-09-09 export (--local-sqlite instead of --db, no network/wrangler involved): 2,016,523 nodes (814,344 entities, 770,176 officers, 26,768 intermediaries, 402,246 addresses, 2,989 other; 1,139 officer/intermediary duplicates deduped), 3,339,267 relationships, 1,611,197-row FTS index, loaded in well under a minute on a single core. The resulting SQLite file is ~1.1 GB — comfortably inside D1's 10 GB (Workers Paid) database-size cap. The production load is dominated by network round trips rather than CPU: the loader packs many byte-capped INSERT statements per file (D1's single-statement cap is 100,000 bytes; wrangler d1 execute --file supports files up to 5 GiB), which works out to ~74 files for offshore_node, ~50 for offshore_relationship, and 41 server-side batches to rebuild offshore_name_fts — roughly 165 wrangler d1 execute round trips total, estimated 45–90 minutes depending on D1 import latency per file.

Quick Start

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

{
  "mcpServers": {
    "icij-offshore-leaks": {
      "url": "https://gateway.pipeworx.io/icij-offshore-leaks/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/icij-offshore-leaks/mcp returns the tools in the table above plus the shared Pipeworx meta-tools — ask_pipeworx, discover_tools, search_within, remember/recall and the rest of the gateway-wide set. So the tool count you see is larger than this table: a single-pack endpoint currently lists roughly 30 shared tools alongside the pack's own. The connection's initialize response states its exact scope, and is the authoritative answer for a given day.

This is deliberate, not multiplexing by accident. The meta-tools are what let a scoped connection answer a question this pack does not cover — via ask_pipeworx, which routes across the whole catalog — without you adding a second MCP server. There is currently no way to mount a pack endpoint without them; if the extra schemas cost you more context than the routing is worth, connect to the full gateway once rather than to several pack endpoints.

Or connect to the full Pipeworx gateway to get every pack's tools listed directly, instead of just this one's:

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

Both URLs reach the same gateway and the same 1807+ data sources. The only difference is which pack's tools are listed directly; ask_pipeworx reaches all of them from either one.

No MCP client? Call it over HTTP

curl -X POST https://gateway.pipeworx.io/v1/tools/offshore_search \
  -H 'Content-Type: application/json' \
  -d '{"name":"Wintris"}'

No account needed for the first calls. Inspect any tool: GET https://gateway.pipeworx.io/v1/tools/offshore_search. Find one: POST https://gateway.pipeworx.io/v1/tools/search_packs with {"query":"..."}.

Standalone (no gateway account)

This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:

{
  "mcpServers": {
    "icij-offshore-leaks": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-icij-offshore-leaks"]
    }
  }
}

Or run it directly to confirm it starts:

npx -y @pipeworx/mcp-icij-offshore-leaks

It speaks MCP over stdin/stdout and answers initialize/tools/list/tools/call for only this pack's tools — none of the shared meta-tools the gateway connection above adds. Same source, same tools, no ask_pipeworx routing.

Using with ask_pipeworx

Instead of calling tools directly, you can ask questions in plain English — this works on the pack endpoint above as well as on the full gateway:

ask_pipeworx({ question: "your question about Icij Offshore Leaks data" })

The gateway picks the right tool and fills the arguments automatically.

More

• Docs and guides
• pipeworx.io

License

MIT