How to screen AI agent payments for OFAC

A practical walkthrough for adding OFAC screening to an AI agent's payment flow before funds move.

TL;DR

TL;DR: Screen agent payments by placing a counterparty check before every transfer, then act on ALLOW or BLOCK. Clean means pay; flagged means stop.

Place the check before the transfer

An agent that pays through x402, Base/USDC, or Coinbase AgentKit is moving money without screening, because those integrations do not check sanctions. Add a call to sanctionsai.dev's sanctions_check right before the transfer. It returns ALLOW or BLOCK in under 100 ms, so the agent can decide with no meaningful delay.

Act on the result

Follow the 4-Gate Agent Payment Protocol: SCREEN the counterparty, SCORE it with risk_score when the result is not clearly clean, STOP on a BLOCK, and STAMP a record of the decision. A clean ALLOW lets the agent pay; a BLOCK halts the payment and logs the counterparty for review.

Add screening as a tool

For frameworks like LangChain, CrewAI, Eliza, or the OpenAI Agents SDK, expose screening as a tool the agent calls before the payment tool. The MCP server, installed with pip install sanctions-mcp, makes that wiring straightforward. The data set covers 19,218 SDN names and 947 OFAC-listed crypto wallets, synced hourly.