An extension pattern that lets MCP-connected tools expose interactive application UI alongside model-visible capabilities.

Constrained model outputs that conform to a defined schema, making AI decisions easier to integrate into software.

The capability for a model or agent to select and invoke external functions, APIs, or tools as part of a workflow.

Infrastructure for giving agents temporary, task-scoped access to external tools and services.

Ephemeral isolated compute for safely running untrusted code generated or used by agents.