Ask most AI assistants about your mainframe data and you get nothing useful. The records are there — customer files, order history, transactions going back decades — but they live in VSAM, described by copybooks, reachable only through code somebody wrote a long time ago.
In this three-minute demo, Senior Systems Engineer Demos Economacos shows what changes when an AI assistant can query that data directly.
The groundwork is data virtualization: in the CONNX Data Dictionary, VSAM record layouts appear as ordinary SQL tables — CUSTOMERS_VSAM, ORDERS_VSAM, PRODUCTS_VSAM — each field’s native type, offset, length, and codepage resolved into standard SQL types. The CONNX MCP Server then publishes those tables to Claude over the open Model Context Protocol: query tools, metadata services, and schema access, all delivered through the security and governance CONNX already enforces.
The rest is a conversation. “How many customers do we have?” returns a count read live from VSAM. “What is our best-selling product?” is harder, and it’s the part worth watching. The assistant has never seen this schema. It inspects the tables, writes its own SQL, gets a query wrong and corrects itself, then aggregates orders by product and answers twice over — one product leads on units sold, a different one leads on revenue. A distinction nobody asked for, and most people would want.
No extract, no replication, no overnight staging. The queries run against the mainframe in real time, and the data never leaves the governed path.
The demo runs against sample data on a VSAM source; the same dictionary shows Db2 and Kafka connections alongside it.