Manage agents, tools and skills at scale with AWS Agent Registry
Amazon Web Services Chaitra Mathur
AWS launched Agent Registry, a shared catalog for agents, tools and skills. It matters because teams can finally find, review and reuse what’s already there instead of rebuilding it.
Based on reporting by Amazon Web Services, Chaitra Mathur — 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 has put Agent Registry into general availability, aiming at a problem that shows up once agentic AI stops being a lab exercise and starts spreading across a company: nobody can keep track of what exists, who owns it, or whether it was ever reviewed. The pitch is blunt. Discovery and governance, not just building, have become the bottleneck.
The registry is meant to be a single catalog for agents, tools, skills and custom resources. AWS splits it into two planes. The Governance Plane is the authoritative store, where admins can attach security and compliance signals, set discovery rules, and define custom metadata such as cost center, data classification or SLA tier. The Discovery Plane is what everyday users and systems see: only approved entries, searchable at scale, with semantic search for intent and lexical search for exact names.
That separation does the real work here. Admins keep full visibility and policy control. Consumers get a faster, cleaner view that hides drafts, rejected entries and shadow resources. AWS says the discovery side is built for high-throughput queries so agents and developers can search programmatically without rate limits getting in the way.
Registry supports four record types: MCP servers, A2A agent cards, skill definitions in markdown with associated code or packages, and custom JSON descriptors. In other words, it is trying to become the system of record for the messy middle of agent infrastructure, not just a directory with a nicer front end.
AWS is leaning hard on early customer stories to make the case. Southwest Airlines says it moved from dozens of scattered agents and tools to a single governed catalog. Sony is using it to reuse agent patterns across business units. Mitsubishi Electric wants one place to discover and trust what developers build. And partners are already wiring in security, data and lifecycle tooling, which is usually how an AWS platform goes from feature to ecosystem.
The broader bet is clear: once enterprises start deploying hundreds or thousands of agents, the winning product won’t just be the one that builds them fastest. It will be the one that can answer a much duller question: what do we have, who approved it, and can anyone trust it?
My take — AI-written commentary, not fact-checked reporting
This is the kind of boring infrastructure that usually wins. Everyone loves agent demos until the second team builds the same thing and no one can tell which version is safe to use. A registry won’t stop the hype machine, but it will put a leash on the mess, which is more useful than another talking chatbot with confidence issues.
Read more about this at: Amazon Web Services
