AWS Backs ARD Open Spec to Federate AI-Agent Discovery Across Environments
The Apache 2.0 specification is designed to make agents, MCP servers, tools and APIs discoverable across separate catalogs without moving their governance into one central system.
Listen to this story
The audio brief
Story brief
3 key pointsAmazon Web Services is backing Agentic Resource Discovery (ARD), an Apache-2.0 open specification for federating discovery across independently governed registries. ARD standardizes resource descriptions and lookup, but leaves authorization, approval, visibility, and revocation with each registry; it is neither a central catalog nor a shared permissions layer. AWS’s Agent Registry, currently in preview, provides the...
- 01
AWS Agent Registry catalogs agents, MCP servers, tools, agent skills, and custom resources.
- 02
ARD does not replace local authorization, approval workflows, or access revocation.
- 03
Agent Registry combines semantic and keyword search and exposes a remote MCP endpoint.
As companies spread AI agents and their tools across clouds, SaaS products and internal systems, finding an approved resource can become a manual integration job. AWS is backing Agentic Resource Discovery, or ARD, an open specification intended to let those separate registries expose and discover resources through a common protocol instead of requiring custom connections between every pair.
ARD is not a product or a central registry. AWS describes it as an interoperability layer that complements AWS Agent Registry, its own centralized catalog for agents, Model Context Protocol servers, tools, agent skills and custom resources. Agent Registry is currently in preview.
A common description replaces pairwise plumbing
The problem ARD addresses is format mismatch. A registry in one environment may use different names and metadata from a registry elsewhere. Connecting each one to another can require a bespoke bilateral connector. ARD defines a shared resource format and discovery protocol, so catalogs that implement it can federate rather than negotiate those one-off links.
The registry remains the gatekeeper
AWS’s companion registry illustrates the division of labor it is proposing. An administrator creates a registry and sets authorization and approval rules. Publishers submit records describing resources; curators can approve, reject or deprecate them; people or agents then search the approved catalog. AWS says a registry can use IAM credentials or a JSON Web Token from a corporate identity provider for authorization.
What AWS Agent Registry adds inside AWS
- Curation and approval workflows intended to keep only records meeting an organization’s security, compliance and quality requirements discoverable.
- Hybrid search that combines semantic matching with keyword lookups.
- A remote MCP endpoint, allowing MCP-compatible clients to search and use the registry directly.
Federation without surrendering local control
AWS’s intended model is that each organization keeps authority over its own catalog: what it publishes, who can see it and when access is revoked. The company expects existing Agent Registry controls to remain the enforcement point, while ARD handles interoperability. It also expects the protocol to create cross-organizational discovery paths when an organization publishes a catalog on its own domain.
That boundary is important. ARD can give participating registries a common way to describe and locate resources, but it is not a shared authorization system or a migration path into a single catalog. Its practical reach will depend on registries and clients adopting the specification. AWS said it contributed feedback during ARD’s development; the specification is available under the Apache License 2.0 through its website and GitHub.
Sources
- aws.amazon.comAgentic Resource Discovery (ARD): An open specification for agent discovery | Amazon Web Services