Agentic Resource Discovery (ARD): An open specification for agent discovery
Amazon Web Services Jeffrey Damick
AWS has a new registry for AI agents and tools, plus an open spec to help them be found across systems. The big deal: it’s trying to make agent discovery work beyond one cloud or one vendor.
Based on reporting by Amazon Web Services, Jeffrey Damick — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AWS is trying to solve a messy problem before it gets worse: once companies have more than a few AI agents, tools, and MCP servers, finding the right one turns into a scavenger hunt. Teams end up wiring things up by hand, and a setup that works in one AI client doesn’t automatically show up in another. That’s fine at small scale. It gets ugly fast after that.
The company’s answer is AWS Agent Registry, a centralized catalog for agents, MCP servers, tools, agent skills, and custom resources. The registry is built around two pieces: registries and registry records. A registry lives in an AWS account and comes with its own authorization and approval settings. A record describes a resource, including what it is, what it does, and how to reach it. Administrators create the registry, publishers submit records, curators approve or reject them, and consumers search for what they need.
AWS is pushing the registry as enterprise-ready because it bakes in curation, hybrid search, MCP-native access, and flexible authorization. The approval flow is there so only resources that clear a security, compliance, and quality bar are discoverable. Search mixes semantic understanding with keyword matching. And because the registry sits behind a remote MCP endpoint, any MCP-compatible client can use it directly.
But AWS also knows the bigger problem isn’t just finding things inside one environment. Enterprises spread agents and tools across clouds, on-premises systems, SaaS platforms, and internal applications, each with its own naming scheme and metadata. That’s where Agentic Resource Discovery, or ARD, comes in. ARD is an open standard under Apache License 2.0, available at agenticresourcediscovery.org and on GitHub. AWS says it should let separate registries describe resources in the same format and expose discovery through a common protocol, so they can federate without proprietary connectors or bilateral deals.
The company is framing ARD as a complement to its own registry, not a replacement. The pitch is simple: keep control local, but make discovery portable. That’s the right direction. The industry has spent years pretending every platform will happily speak every other platform’s dialect, and somehow all that produced was more glue code. Open discovery is less glamorous than a shiny agent demo, but it’s the part that might actually scale.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI infrastructure story that sounds boring and matters a lot. The sector keeps shipping more agents than anyone can track, then acts surprised when discovery turns into a vendor-specific mess. Open standards like ARD are what stop the whole thing from becoming a pile of secret handshakes with better branding.
Read more about this at: Amazon Web Services