GraphQL and MCP as structured architecture for enterprise AI agents

At the AI Agent Conference, Apollo GraphQL CEO Matt DeBerglis sat down with Ryan for a recorded conversation about the architectural challenges enterprises face when deploying autonomous AI agents. The central argument is that GraphQL and MCP, used together as a structured semantic layer, can give agents access to clean, well-defined data rather than unstructured or overly broad context — directly addressing both the quality of agent reasoning and the cost of running it.

DeBerglis raised two specific enterprise concerns. The first is a security risk he describes as "east-west" data exfiltration — the threat that agents moving laterally across internal microservices could inadvertently or maliciously expose sensitive data in ways traditional perimeter security does not anticipate. The second is token spend: by explicitly querying only the exact context an agent needs, rather than flooding the context window, enterprises can rein in the rapidly escalating costs associated with large-scale agent deployments.

Key facts

1. 01Recorded live at the AI Agent Conference and published on the Stack Overflow Blog by Phoebe Sajor.
2. 02Apollo GraphQL CEO Matt DeBerglis is the featured guest, interviewed by Ryan.
3. 03The discussion centers on using GraphQL and MCP as a structured semantic architecture for autonomous agents.
4. 04DeBerglis identifies "east-west" data exfiltration as a new security risk agents pose to internal microservices.
5. 05Explicitly querying only the exact context required is presented as the solution to skyrocketing token spend.

Topics