Memgraph MCP and Claude Desktop: Analyzing Test Banking Data for Mule Accounts (andrewbaker.ninja)

A new how-to demonstrates running Memgraph locally with Claude Desktop via the Model Context Protocol (MCP) to analyze banking transaction graphs and surface mule-account networks. The provided automated script pulls memgraph/memgraph-platform in Docker (exposing ports 7687/7444/3000), installs a lightweight MCP server using the uv runner, configures Claude Desktop to call an mcp-memgraph adapter, and seeds a test dataset (57 accounts, 512 transactions) that hides a realistic “hub-and-spoke” mule laundering topology. The setup also creates indexes (e.g., :Account(account_id), :Person(person_id)) and exposes mgconsole for direct Cypher usage; all communication is local so data does not leave the machine. Significance for AI/ML practitioners is twofold: graph databases make multi-hop relationship queries, pattern matching, and real-time traversal trivial compared to complex recursive SQL, while Claude Desktop provides a natural-language layer that translates analyst questions into optimized Cypher—lowering the barrier for feature exploration, labeling, and rapid hypothesis testing. Technically this means analysts can quickly extract paths, subgraph motifs, or neighborhood features for downstream ML (GNNs, anomaly detection), iterate on detection rules interactively, and prototype production workflows where LLMs orchestrate queries against a fast, in-memory graph backend.