AWS has launched Agent Registry in preview, giving enterprises a central catalogue for AI agents, tools, and related resources as organisations move from isolated pilots to broader deployment.
Available through Amazon Bedrock AgentCore, the service is designed to help platform and IT teams discover, approve, and reuse agents across AWS, other cloud providers, and on-premises environments. AWS says the registry stores structured metadata on agents, tools, MCP servers, agent skills, and custom resources, including ownership, capabilities, invocation methods, and compliance details.
The launch is aimed squarely at a growing governance problem inside large organisations. As more business units build agents for internal workflows, support, finance, software delivery, and customer-facing tasks, central technology teams are being left to track what exists, who owns it, whether it meets policy requirements, and whether another team has already built something similar. AWS says the registry is intended to reduce that duplication while giving administrators a stronger approval process and clearer audit trail.
According to AWS, records can be added manually through the console, SDK, or API, or pulled in automatically from MCP or A2A endpoints. The company says the registry is accessible through the AgentCore console, APIs, and as an MCP server, which means both developers and MCP-compatible clients can query it directly. For organisations using their own identity systems, AWS also says OAuth-based access allows teams to build their own discovery interfaces without depending on IAM credentials for every use case.
“Discovery becomes the path of least resistance,” AWS says. “Teams can search by name, descriptions, and resource type to find what already exists before building something new.”
AWS’s own documentation says Agent Registry combines semantic and keyword search, supports approval workflows, and can be organised by team, business unit, development stage, or resource type. Administrators can set authorisation rules, reviewers can approve or reject records, and older entries can be deprecated when they are no longer in use. That matters for enterprises trying to stop agent estates becoming yet another unmanaged layer of software inventory.
The timing is notable. Enterprise AI programmes have moved beyond basic experimentation and into questions of operating model, policy, and cost control. The issue for CIOs is no longer just how to build an agent, but how to know which agents exist, how they connect to tools and data, which ones are approved, and whether multiple teams are solving the same problem in parallel. In that context, a registry becomes less a developer convenience and more a control mechanism for governance, reuse, and standardisation.
AWS said the preview is available in five AWS Regions — US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), Asia Pacific (Tokyo), and Europe (Ireland). The company has also published documentation for Agent Registry. The release does not remove the need for internal governance, but it does give enterprise teams another managed layer for cataloguing AI assets, enforcing approval steps, and reducing duplication as agent adoption scales.
