MCP-ization and GraphQL | Roy Derks (Principal Product Manager, IBM)

Listen to full episode:

Summary: In this episode, Anant, Ed, and guest Roy Derks (Principal Product Manager at IBM) unpack why Model Context Protocol (MCP) is becoming a foundational layer for agentic applications—and why GraphQL may be the best way to curate and expose the data behind it. The conversation moves from API history to practical developer workflows, showing how curated, well-structured interfaces reduce tool sprawl, improve reliability, and make AI systems more useful.

Chapters:

00:00 - Reconnecting and introducing Roy Derks

01:01 - Why MCP is becoming the center of conversation

05:46 - Why enterprise context and inside knowledge matter

08:46 - The API management parallel behind MCP hype

10:30 - REST vs. GraphQL for compositional data access

14:32 - GraphQL and MCP as similar architecture tradeoffs

17:25 - Tool reuse and MCP as the interoperability layer

21:46 - NPM-style reuse and the shift to MCP hub-and-spoke

26:06 - The OpenAPI problem and echoes of GraphQL's origins

31:50 - Speed vs. doing MCP right, and tool overload

36:36 - Roy's demo: REST to GraphQL to MCP to agent loop

43:20 - Curating a smaller MCP surface by combining tools

46:46 - Live demo: natural language to GraphQL and back

51:40 - Final takeaways: GraphQL as the enterprise MCP layer

Sound Bites:

“GraphQL as a language is a declarative language, meaning it's easy to use and read for developers, but thereby also easy to read and use by large language models.”