What changed
On August 24, AWS announced Agentic Resource Discovery (ARD), an Apache-licensed open specification for describing and discovering agents, MCP servers, skills, APIs, and other callable services across registries. The goal is to let a publisher describe a capability once while compatible clients discover it across different environments.
The product side is AWS Agent Registry, now available in preview. It provides a managed catalog for agents, tools, MCP servers, skills, and custom resources. Teams can publish records, route them through approval and curation, then expose approved resources through hybrid keyword and semantic search or a native MCP endpoint.
Why it matters
The hard part of an agent platform is shifting from making one agent call one tool to helping many agents find the right capabilities safely. Without a catalog, teams wire integrations by hand, duplicate tools, and lose track of which endpoints are approved or still maintained.
AWS is separating two layers that are easy to conflate: the Agent Registry is the governed product boundary, while ARD is the interoperability layer. AWS describes ARD as a way for registries in different clouds, on-premises systems, and SaaS products to federate without bilateral connectors. That is a meaningful product move: discovery becomes a managed workflow with ownership, authorization, approval, search, and revocation instead of a pile of client configuration files.
Builder takeaway
If you are building an agent product, treat capability discovery as part of the user experience and the operating model. Give tools and agents durable metadata, explicit versions, ownership, and lifecycle states. Put an approval boundary around what agents can discover, and expose discovery through a protocol that can survive a change of client or hosting environment.
The maturity caveat matters. AWS Agent Registry is still in preview, and the ARD specification repository labels the standard v0.9 and draft. The practical takeaway is not to outsource your architecture immediately; it is to keep registry metadata and discovery behind a clean boundary so a shared standard can be adopted later without rewriting every agent.