AWS launches Agent Registry in preview - a catalog of AI agents and tools for companies

On April 9, 2026, AWS introduced AWS Agent Registry in preview status. This is a component of AgentCore designed to consolidate corporate AI agents, tools, MCP servers, and agent skills in one place, enabling teams to not only find them, but also reuse them without chaos and duplication.

Why a Registry Is Needed

According to AWS's description, the problem is no longer how to build a single agent, but how to manage hundreds and thousands of such systems within a large company. When some teams deploy agents on AWS, others use third-party clouds, and still others keep them on-premises, visibility is quickly lost: no one knows exactly what already exists, who is responsible for it, and whether it can be reused. As a result, identical functions are created multiple times, and both costs and compliance risks grow alongside.

AWS positions Agent Registry as a unified catalog spanning this fragmented environment. The service indexes entries regardless of where an agent was created or where it operates: in AWS infrastructure, with another provider, or within the company's own environment. This is an important emphasis for AWS: the registry is designed not as a showcase for its own services, but as an accounting and management layer for a hybrid corporate stack.

What's Available Now

In preview, the registry stores structured metadata about each agent, tool, MCP server, agent skill, and custom resource. The entry can specify the publisher, supported protocols, available capabilities, and invocation method. AWS also announces native support for MCP and A2A standards. Data can be added manually through the console, SDK, or API, or you can simply specify an MCP or A2A endpoint, after which the service will automatically pull the description. This allows you to populate the catalog with an existing agent landscape rather than starting from scratch.

• Manual registration via console, SDK, and API
• Auto-import of metadata from MCP and A2A endpoints
• Access to the catalog via Console, API, and MCP server
• Hybrid search: by keywords and by semantic meaning
• OAuth access for custom search interfaces without IAM accounts

AWS separately emphasizes making existing components easy to find. Registry search combines conventional keyword matching with semantic matching for long natural language queries. In an example from the announcement, a query about payment processing can find tools described as billing or invoicing, even if the names don't match. The logic is straightforward: a developer first searches for what's already approved within the company, and only then writes a new agent from scratch.

How Governance Works

The second pillar of the service is governance. AWS offers to manage publication and access through IAM policies: administrators can separately define who has the right to register agents and tools, and who can find and use them. Each entry goes through a lifecycle: first draft, then pending approval, and only after approval does the object become visible to the entire organization. Additionally, versions, decommissioning, and arbitrary metadata such as team owner, compliance status, or deployment environment are supported.

In the announcement, AWS provides early corporate cases. Zuora, which has already deployed around 50 agents for Sales, Finance, Product, and Developer teams, uses this approach for a centralized catalog and reuse of ready-made components. Southwest Airlines emphasizes something different: if unified publication and ownership rules are not established from the start, the agent stack quickly becomes an uncontrolled collection of disparate services. Agent Registry attempts to close this governance gap.

What This Means

AWS is clearly building not just another AI service, but an infrastructure layer for an enterprise agent ecosystem. Even now, the company is talking about the next step: automatic agent indexing at deployment, federation of multiple registries, categories and taxonomies tailored to business needs, as well as linking operational metrics from AgentCore Observability. Preview is currently available in only five AWS regions, but the direction is clear: enterprise teams need not only a way to create agents, but also a unified registry where they can search, validate, approve, and decommission them according to rules.