Field noteNov 18, 20266 min read1 source

Govern and discover records with common metadata in AWS Agent Registry

In one minute

AWS says custom metadata schemas are now available in AWS Agent Registry, letting administrators define validated fields for registry records such as agents, MCP servers, and skills. The feature connects a shared vocabulary to approval and discovery, so teams can filter for resources that meet operational and security criteria rather than relying on names or descriptions alone.

What changed: AWS introduces common metadata schemas in Agent Registry to govern and discover records, extending its registry automation capabilities.

What happened

AWS’s announcement says an administrator can define a custom metadata schema for an AWS Agent Registry. The schema specifies field names, types, allowed values, and whether a field is required or optional. Each field can be Text, Enum, URL, or Boolean. A schema includes attributes shared by all record types, with additional fields available for particular types. The registry validates metadata when a record is created or updated. A record created without custom metadata is saved as NON_COMPLIANT; it can remain a work in progress but cannot be submitted for approval until it has the required fields and values. Unknown fields, missing required values, and invalid values trigger a ValidationException. The GetRegistryRecord operation returns both the metadata and compliance status. AWS distinguishes compliance with the schema from approval: a compliant record is not automatically approved.

The walkthrough uses AnyCompany, which operates a registry without auto-approval. Its payments team publishes payments-ledger-mcp, a read-only MCP server for ledger lookups. The schema makes environment, businessUnit, and securityScanStatus required. It allows prod, staging, or dev for the environment, and PENDING, PASSED, or FAILED for the scan status. Other common fields include ownerTeam, cmdbId, dataClassification, and securityScanReportUrl; MCP records can also have a networkEgress field. When the publisher submits production as the environment, the registry rejects it because the allowed value is prod. Correcting the value creates a compliant draft.

Next, the company’s automation adds information from its security scanner and configuration management database, then submits the record for approval. AWS notes that an update replaces the whole metadata map, so the sample first reads the existing metadata, combines it with new values, and writes the complete map back. The scanner and CMDB calls in the example are stubs to be replaced with a team’s own integrations. A pending-approval event goes to Amazon EventBridge; an EventBridge rule invokes a custom AWS Lambda function. That function checks that the record is compliant, the scan passed, a CMDB ID exists, and the data classification is public or internal. The example approves the record if it passes all four checks. AWS says teams can route confidential or restricted records to human review instead.

For discovery, the example filters on environment=prod, businessUnit=payments, and securityScanStatus=PASSED, alongside a natural-language query for a ledger lookup. Only approved records are returned. AWS says agents can make the same search through the registry’s MCP endpoint, which exposes search, list, and get as MCP tools; the sample connects a Strands agent to that endpoint and uses a model on Amazon Bedrock. AWS lists prerequisites of an AWS account in a supported Region, IAM permissions and credentials for Boto3, Python 3.11 or later, and boto3 1.43.113 or later. The announcement points to a sample implementation in the AWS Labs AgentCore samples repository.

What the archive adds

New Runtime’s earlier coverage described AWS AgentCore governance with Okta Cross App Access in an earlier note dated 2026-11-07. That concerned controlling agent access, while this is a new AWS Agent Registry capability for structuring metadata and using it in approval and discovery. It is related governance work, not a restatement of the same event.

Trend signal

This supports, narrowly, agent-ready software exposes capabilities: consistent, filterable metadata makes registry resources easier for agents to discover within organizational constraints. The announcement demonstrates a concrete AWS implementation, but does not establish how broadly teams will adopt this design or whether it will become a common interoperability layer.

Why now

AWS frames the feature around organizations building more agents, MCP servers, and skills, and needing a consistent way to describe, govern, and find them. The timing is therefore tied to managing a growing fleet of registered resources, rather than to a general launch date or deadline described in the announcement.

Why it matters

A registry is useful only if teams can tell what a resource does, who owns it, where it runs, and whether it satisfies the checks required for use. Shared fields give publishers, approvers, consumers, and agents a common vocabulary. More importantly, AWS’s example connects that vocabulary to operational gates: existing systems contribute scan and inventory facts, a function evaluates them, and the same values support subsequent search. This can reduce ambiguous handoffs between the team publishing an MCP server and the team deciding whether to expose it.

The boundary matters. COMPLIANT means that a record matches the schema; it is not a security verdict or an approval. Teams still define the checks, determine which cases need human review, and maintain the systems that supply trustworthy values. AWS’s advice to keep secrets, internal endpoints, and detailed findings out of broadly readable metadata is especially important when the registry serves agents as well as people.

What happens next

  • Teams adopting custom schemas will need migration and review procedures for schema changes. (6 months, 65%). Watch for: AWS guidance or customer examples describing how existing records are kept compliant as required fields evolve.

Limits and uncertainty

AWS presents a worked example, not measured evidence that the feature reduces approval time, improves security outcomes, or scales across large fleets. The example’s scanner, CMDB, and approval logic require customer implementation; the code does not supply those integrations as finished services. AWS also warns that schema changes are additive: fields cannot be renamed, removed, or retyped, and existing records are not rewritten automatically. A record may become NON_COMPLIANT after a schema change while retaining its approval status; its next update must satisfy the new schema. The announcement describes the mechanism, but its operational cost and behavior across different organizations’ controls have not been independently assessed here.

Skill to build now

Draft a registry schema with the teams that publish and approve resources before implementing automation. Put broadly applicable facts in common fields and resource-specific facts in type-specific fields; use enums for decision values that must match exactly. Make a field required at registration only if the publisher can supply it then. For later checks, allow a state such as PENDING, and have enrichment read, merge, and replace the full metadata map. Keep evidence URLs access-controlled, and route sensitive classifications to a human review path.

Reader takeaway

AWS Agent Registry’s custom schemas turn organizational metadata into inputs for both approval and search. The useful pattern is the lifecycle: agree on a vocabulary, enrich records from existing systems, make approval rules explicit, and let agents discover only approved matches. Schema compliance helps enforce consistency, but it does not replace security review or ownership of the data behind each field.

Evidence / sources

  1. 1aws.amazon.com/blogs/machine-learning/govern-and-discover-records-with-common-metadata-in-aws-agent-registry
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicMcpExplore the mcp topic hub.
  2. 02archiveField NotesOpen the latest editorial analysis.
  3. 03sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract