REST is for Agents: Rediscovering Spring HATEOAS
summary
A working adapter called MoneyMate exposes a legacy Open Bank Project sandbox (≈700–800 APIs) as a HAL forms HATEOAS API so agents discover allowed actions at runtime; a lightweight agent Goose drives it with no hardcoded tools while an MCP-based token exchange and Spring Authorization Server handle auth and avoid pasting credentials. Implementation uses Spring HATEOAS, a REST client to map a subset of endpoints, and an HTTP-session MCP shim; result: dynamic, identity-aware tool lists, far smaller context footprints, and enforced server-side workflow.
AI-written summary · about
chapters (11)
- 0:00AI agents and legacy APIs
- 3:13AI agents with tools
- 7:28MCP limitations for APIs
- 11:17HATEOAS for agent navigation
- 13:22Goose agent demo
- 18:43HATEOAS API with Goose
- 28:44Spring HATEOAS adapter
- 30:30Authentication with MCP and OAuth
- 38:18MCP vs HATEOAS
- 39:44Custom agentic loops vs frameworks
- 41:50Key takeaway: agents as systems
related talks
21mBuilding Flexible APIs with Spring HATEOAS
46mGet HATEOAS and Hypermedia right with Spring
47mHypermedia IRL
57mSecuring RESTful services with Spring HATEOAS and HDIV
51mREST beyond the obvious – API design for ever evolving systems
01h 35mSpring Data REST - Data Meets Hypermedia
46mREST next level: writing business-oriented web APIs
01h 31mCreating REST-ful, Hypermedia-based Micro-services with Spring Boot