# AWS Agent Registry Public Search: How to Review a Tool Before You Connect It

Use AWS Agent Registry Public Search to check publisher identity, review tool metadata, and keep external tools separate from your approved catalog.

By Paras Madan | 2026-11-22


If someone on your team finds a tool in a public agent registry, do you want the tool to show up as a candidate or do you want an agent to start using it? Those are two different decisions. The difference matters when the tool asks for access to a real workspace.

AWS published the [Agent Registry Public Search API reference](https://docs.aws.amazon.com/agent-discovery/latest/APIReference/Welcome.html) on November 20. The new `SearchPublicRecords` operation gives AWS-authenticated callers a way to find agents, tools, and skills that publishers made discoverable outside their organization. I would use it to shortlist tools for review and keep my agent's execution list unchanged.

## Use the right search surface

[Public search](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/registry-public-search.html) and enterprise search are not two views over the same list. Enterprise search returns records approved for use in your organization's registry. Public search returns records promoted beyond that organization. It requires AWS authentication; public does not mean anonymous.

In the AWS Agent Registry console, open **Public discovery** and search for a capability you actually need. AWS lists public search endpoints in Northern Virginia, Ireland, and Tokyo. If a result you're expecting is missing, first check which search surface you're on. Internal approval alone won't put a record into public results.

## Check the publisher and the exact record

Suppose you need a tool that reads the status of a test support ticket. Search for that job, then open the candidate's details. Record the publisher, record identifier, and version from the listing before anyone connects it. The [search operation reference](https://docs.aws.amazon.com/agent-discovery/latest/APIReference/API_SearchPublicRecords.html) documents filters such as record type and domain that can narrow a programmatic search. That helps when several tools have similar names. It doesn't turn the first match into an approved choice.

Check the publisher against the vendor's own documentation. A verified publisher gives you an identifier to check; it doesn't tell you whether a tool needs more access than your task warrants. Look for the exact operation, the authorization it needs, and what data it sends outside your workspace. If the public descriptor is too thin to answer those questions, ask the publisher for documentation rather than inferring the answers from the search result.

## Make the first connection a separate decision

For that ticket tool, I'd use a synthetic ticket in a test workspace and authorize only the read operations needed for the check. Compare the returned status against the ticket system. Try a ticket the test account shouldn't be able to see and confirm the access is denied. Check that the tool you evaluated has no permission to edit or send anything under the test credentials. A successful answer on the first ticket says little about the other two checks.

Record the version you reviewed and who approved it. If the endpoint, requested permissions, or published revision changes, review the change before letting agents keep using it. The results list can point you toward tools, but it doesn't need to double as the list of tools your agent is allowed to run.

## If you're the publisher, preview what outsiders will see

AWS's [public promotion guide](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/registry-promote-to-public.html) separates approval of the registry record from approval of its association with a verified publisher. A record approved only for internal use stays out of public search. Once it is promoted, other AWS-authenticated callers can see the public metadata; encryption at rest does not hide it after publication.

Review the description and examples for employee names, private hostnames, and internal sample data before approving the association. Then check the listing from an independently authenticated AWS account you control. You want to see the intended publisher and version without revealing where an internal test system lives. If you're on the buying side, do the reverse: treat that listing as the start of the review, not the end of it.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI infrastructure teams explain how their products fit into real workflows through clear technical writing and practical storytelling.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
