The future of large-scale agent management: AWS Agent Registry is now in preview mode.

Today we are announcing AWS Agent Registry (preview) in AgentCore — a single place for discovering, sharing, and reusing AI agents, tools, and agent skills within your organization.

As the number of agents in organizations grows to hundreds and thousands, platform teams face three key challenges: visibility (awareness of existing agents), control (management of publishing rights and information availability), and reuse (preventing duplication of already existing capabilities).

The lack of a centralized system accelerates the proliferation of agents, increases compliance risks, and leads to wasted efforts on duplicating work. These issues are exacerbated by the fact that no organization is limited to a single service provider: agents are created on different platforms, including AWS, other cloud services, and on-premises environments.

A registry that covers only part of the stack leaves the other components invisible, making it impossible to discover, manage, or reuse them. To address this challenge, more is needed than just a list of existing agents. Platform teams must be able to create agents, publish them with approval processes, and help other teams find and reuse existing ones.

In this regard, we introduce AWS Agent Registry (preview) in AgentCore — a unified platform for discovering, sharing, and reusing AI agents, tools, and agent skills within your company. AgentCore is built with a focus on agents, providing the ability to work with any models, frameworks, and enterprise architectures.

Whether you are developing your first agent or your thousandth, you get a platform that grows with you. The registry provides the same flexibility in organizing and managing what has been created. It indexes agents, regardless of where they are developed or hosted — on AWS, other cloud providers, or in on-premises environments.

Currently, in preview mode, the registry stores metadata for each agent, tool, MCP server, agent skill, and custom resources in a structured format. It captures who published each entry, what protocols it implements, what it offers, and how it can be invoked.