I built an MCP server that screens suppliers against 328 sanctions lists before your agent approves a payment

#ai #mcp #agents #kyc

I've been building niche MCP servers for AI agents. This one solves a problem I kept running into: agents that approve payments or onboard suppliers with zero counterparty verification.

The risk is real. A single payment to a sanctioned entity creates criminal liability. A single fraudulent supplier invoice with no registry check creates unrecoverable financial exposure. Agents don't know this — they just approve.

What I built

Bizfile MCP is a Model Context Protocol server with three tools:

validate_counterparty — checks UK Companies House, Singapore ACRA, and OpenCorporates (210M+ companies across 130+ jurisdictions). Returns company status, directors, filing history, AI risk score 0-100, KYC confidence HIGH/MEDIUM/LOW, and a PROCEED/EDD/BLOCK verdict. One call, machine-ready output.

screen_counterparty — screens against 328 global sanctions lists including UN, EU, OFAC, UK HMT, and MAS Singapore via OpenSanctions, updated daily. Returns CLEAR/MATCH verdict and matched list names.

validate_counterparty_lite — fast registry-only check for high-volume workflows. No AI scoring, sub-second response, lower token cost.

Example

curl -X POST https://bizfile-mcp-production.up.railway.app \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"validate_counterparty","arguments":{"company_name":"ROLLS-ROYCE PLC","country":"GB"}}}'

Returns:

{
"agent_action": "PROCEED",
"registered_name": "ROLLS-ROYCE PLC",
"registration_number": "01003142",
"status": "active",
"kyc_confidence": "HIGH",
"risk_score": 12,
"risk_level": "LOW"
}