MCP Apps: Give the Model Data, Give the User a UI
Most MCP tools return text. MCP Apps let you go further. But the real unlock isn't just rendering a pretty UI, it’s understanding that the model and the user need fundamentally different things from the same interaction. This talk presents a design pattern for building great MCP Apps: separate the data layer (what the model reasons about) from the display layer (what the user interacts with). When you do this well, the model retains full context and agency over structured data, while the user gets a rich, interactive interface. We'll walk through concrete examples of how splitting data and display unlocks capabilities that pure UI apps can't provide: letting the model make choices around display, answer questions based on interactions, and providing detailed displays and filters. Attendees will leave with a practical mental model for designing MCP Apps that are good for both the human and the AI. Attendees will learn patterns they can apply immediately.
About the Context Engineering Track
Context Engineering sessions at AI Engineer World's Fair 2026 in San Francisco.
When
Wednesday, July 1, 2026
2:25 PM - 2:45 PM·20m
Where
Track 8 · Room 2020
Capacity: 250 attendees
Sign in to add this talk to your schedule.
Speaker
Dustin Mihalik
Software Engineer, AI Platforms
Indeed
Dustin is a Technical Fellow at Indeed, where he leads the AI Platform powering all of Indeed's production AI products. He has spent years building the foundational infrastructure that enables engineering teams to develop, deploy, and scale AI applications across the company — from early prototypes to high-throughput production systems.