Building an Agentic Fraud Investigation System with TigerGraph, LangGraph & MCP

A 9-node LangGraph agent was developed to investigate fraud cases end-to-end, utilizing TigerGraph Community Edition as the knowledge graph substrate. The system integrates TigerGraph MCP for tool access and NVIDIA NIM for LLM reasoning. It successfully processed 20 HHGOA benchmark cases, generating FinCEN-standard SAR narratives and recording results back to the graph. This approach addresses the limitations of traditional detection systems and the risks of granting raw LLMs direct access to core banking APIs.