AWS makes Agent Registry generally available for governed agent discovery

Amazon Web Services has made AWS Agent Registry generally available, introducing a centralized catalog for organizations managing growing inventories of AI agents, tools, skills, and related resources.

The service is aimed at a problem that emerges after teams move beyond isolated experiments: capabilities are built by different groups, ownership becomes unclear, and developers may recreate tools that already exist elsewhere. AWS says Agent Registry is designed to make those resources searchable while adding approval, access, and lifecycle controls.

The announcement comes from the AWS Machine Learning Blog, making AWS the primary source for the product’s availability and capabilities. The available evidence does not include independent customer adoption data or third-party performance testing, so the strongest claims about enterprise value remain vendor-reported.

A catalog for an expanding agent estate

AWS describes Agent Registry as a single, searchable, governed catalog for resources across an organization. Teams can register agents, tools, skills, and custom resources, while administrators maintain information about ownership, status, review history, and access.

The catalog is intended to cover several emerging agent formats. AWS identifies Model Context Protocol servers, along with their tools, resources, and prompts, as one supported resource category. It also references Agent2Agent agent cards, which describe agents and their skills, as well as skills represented by Markdown files and associated code or packages.

That scope matters because enterprise agent systems rarely consist of one model or one application. A production workflow may call an internal tool through MCP, delegate a task to another agent, and rely on a reusable skill package. Without a shared inventory, each integration can create another local description of the same capability.

AWS says the Registry is intended to reduce that duplication by giving teams a common place to publish and find what is already available. The company also positions it as a control point for resources that are still under review, rather than only a directory of approved components.

Separating governance from discovery

A central design choice in AWS Agent Registry is the separation between a Governance Plane and a Discovery Plane.

The Governance Plane is the comprehensive store for registered resources within an organization’s defined scope. Administrators can attach compliance and security signals, create organization-specific metadata schemas, and set discovery policies based on entitlements. Example metadata fields include cost center, data classification, and service-level agreement tier.

This plane is designed to preserve visibility into resources regardless of whether they are approved, rejected, in draft form, or no longer active. That distinction is important for security and operations teams: a resource that should not be available to general developers may still need to remain recorded for review, ownership, or audit purposes.

The Discovery Plane is the consumer-facing view. AWS says only resources that have cleared an organization’s approval process appear there, allowing developers and agents to search a curated catalog rather than the full set of registered records.

AWS says Discovery Plane searches can combine semantic search with lexical search. In practical terms, a developer could search by intent, such as looking for a tool that handles ticket routing, or use an exact resource name. The service is also described as supporting programmatic, high-throughput queries for agents and developers.

The consumer view is meant to expose summarized trust signals rather than the full set of governance details. That could help a developer decide whether to use a resource without requiring access to internal compliance records, although the available announcement does not specify the complete set of signals or how organizations must configure them.

What AWS has—and has not—demonstrated

The confirmed product news is availability: AWS says Agent Registry is now generally available and documents publishing, curation, and discovery workflows. AWS also describes access control, lifecycle tracking, approval workflows, semantic search, and custom metadata as core parts of the service.

However, the source is an AWS product blog rather than an independent evaluation. It provides no verified figures for search latency, query capacity, registry size, cost reduction, reuse rates, or the number of customers using the service. AWS’s statements that the Registry can support high-throughput queries and alleviate duplication should therefore be treated as vendor claims until customers or external benchmarks provide more evidence.

The post also distinguishes between current functionality and planned development. AWS says richer governance signals will be surfaced over time, and notes that some described features are forward-looking. Buyers will need to confirm which controls are available in their region and service configuration before treating the Registry as a complete compliance system.

Why it matters to builders and enterprises

For AI builders, the immediate value is less about generating a new agent and more about making existing capabilities reusable. A searchable catalog can give developers a way to find approved tools before writing new integrations, while ownership and lifecycle records can make it easier to identify an outdated or unsupported dependency.

For enterprise platform teams, the two-plane model addresses a tension that often appears in internal marketplaces. Developers need fast discovery, but security teams need a complete record that includes resources consumers should not see. Keeping those views separate could allow organizations to preserve administrative oversight without exposing unapproved agents or tools in everyday search results.

The governance model may also affect deployment reliability. If resources carry version, ownership, security, and classification metadata, teams have more information when tracing a failure or deciding whether a capability is suitable for a sensitive workflow. That does not guarantee that an agent will behave safely, and a registry cannot replace testing, identity controls, or runtime monitoring. It can, however, provide an inventory layer that those processes can reference.

The competitive significance is broader than AWS’s individual service. As companies assemble systems from AI agents, MCP tools, A2A agents, and reusable skills, discovery and policy management become infrastructure concerns. AWS is positioning Agent Registry as that layer within its cloud platform, but the announcement does not establish whether customers will use it as an organization-wide system of record or primarily for AWS-centered deployments.

What to watch next

The next signals will be customer evidence and product detail. Buyers should look for documented pricing, supported regions, API and integration coverage, and clearer explanations of how Registry metadata connects with identity, logging, security review, and runtime execution.

It will also be important to see whether AWS publishes evidence of reuse across teams, changes in duplicated development work, or operational outcomes from the service. Further details on planned governance signals could show whether the Registry becomes a basic catalog or a deeper control layer for enterprise AI.

Creati.ai perspective

AWS Agent Registry targets a real operational bottleneck: once an organization has many agents and tools, knowing what exists and what can be trusted may matter as much as building another capability. The separation of administrative records from approved discovery is a sensible response to that problem.

But a catalog is only as useful as its metadata, approval discipline, and adoption. The general availability announcement establishes AWS’s product direction, not yet its market impact. Enterprise teams should evaluate whether the Registry fits their broader agent architecture, especially across non-AWS systems, and demand evidence that governance improves reuse and reliability rather than adding another inventory to maintain.