Using a sanctions MCP server: how an AI assistant can check OFAC data

What MCP is, in one paragraph

MCP (Model Context Protocol) is an open protocol for connecting an AI assistant to outside data and tools. The protocol defines a tool list and a call format; a client (Claude Code, Claude desktop, ChatGPT or an SDK) can discover the tools a server offers and call them during a conversation. A sanctions MCP server returns authoritative sanctions data in that format, so the assistant can quote the official record instead of what it remembers from training. For the companion piece on the record tools, see OFAC recent actions and sanctions screening in your AI assistant.

Connect in two minutes

Claude Code (command line):

claude mcp add --transport http distributed-risk https://www.distributedrisk.com/mcp

Clients that read a config file (Claude desktop, custom hosts):

{
  "mcpServers": {
    "distributed-risk": {
      "type": "http",
      "url": "https://www.distributedrisk.com/mcp"
    }
  }
}

The server speaks Streamable HTTP (MCP protocol version 2026-07-28), needs no key and makes no writes. One address can make about 120 calls a minute. The full tool list is on the AI access page.

The sanctions tools, by task

Task	Tool	What the assistant gets
Look up a name, alias or identifier	`search_screening_lists`	OFAC SDN, non-SDN Treasury lists, BIS (Entity, DPL, Unverified, MEU), State ITAR-debarred and nonproliferation. Close spellings match. Each result links the official entry, with a last-changed date
Look up a digital-currency address	`search_screening_lists` with an address	OFAC digital-currency SDN entries, with the symbol and program
See what OFAC did recently	`list_ofac_actions`	Designations and removals, general licences, FAQs, regulations, settlements, newest first
Find a Federal Register rule or notice	`search_federal_register`	OFAC, BIS, State rules, proposed rules, notices and presidential documents, by term, agency or date
Pull a desk note with its reasoning	`list_notes`, `search_notes`, `get_note`	The desk's dated notes with numbered sources, as a reader of /desk sees them

What to type at the assistant

"Search the US screening lists for the name 'Example Trading LLC' and return anything matching. Say when each list last changed." → search_screening_lists.
"List OFAC's actions since 1 October 2026." → list_ofac_actions.
"Does OFAC's SDN list include this digital-currency address?" → search_screening_lists with the address.
"Pull the Distributed Risk desk notes on Venezuela this month, with sources." → list_notes or search_notes, then get_note.

The tools return source links in every answer. The assistant can quote the entry, say when the list was last changed and show the OFAC URL.

What this does not do

A sanctions MCP server is a research aid built on the public record. It does not:

Decide whether a transaction is permitted. Counsel and a sanctions programme do.
Replace a formal screening system. Readers own their compliance obligations.
Give an exhaustive map of blocked crypto wallets. OFAC states its digital-currency listings are not likely to be exhaustive. See Checking a crypto wallet or exchange for sanctions exposure.
Apply the 50 Percent Rule for you. The rule turns on beneficial ownership that is not always on the list. See How to screen a counterparty against OFAC lists, and what a match means.

A list match is a lead to check against the official entry, not a determination. No match is not a clearance.

When the record is not enough

For transactions that need people on the ground — a diligence visit, an in-country investigation, protection on a trip, or insurance — brief an advisor: we arrange it with a partner firm that works there.

Questions

Do I need an OFAC API key? No. The Distributed Risk MCP server reads OFAC's and State's public pages and the Consolidated Screening List data file, and makes no writes.

Does this work with ChatGPT? Yes, with any client that supports remote MCP over Streamable HTTP. Current clients include Claude Code and Claude desktop; others are listed on modelcontextprotocol.io.

How current is the data? The sanctions sources are read again on a short cadence: OFAC Recent Actions every 20 minutes, the Consolidated Screening List every hour. The tool answer says when each list was last read.