An operations agent diagnosing a production incident shouldn't need to know which internal system owns the logs, which team wrote the runbook, or which partner's agent can fix a DNS configuration. Today, it does, because every team's tools sit behind a custom registry that nobody else talks to. The Agentic Resource Discovery (ARD) specification, published by Google with industry partners, tears down those walls by giving agents a standard way to find, verify, and connect to any capability published under an organization's own domain.
ARD answers three questions no agent ecosystem has solved consistently: where is the right capability, which one should I use, and how do I verify it's safe to connect? Rather than forcing a single protocol or framework, ARD defines two primitives - catalogs and registries - that layer discovery and trust on top of whatever native protocol the tool already speaks (MCP, A2A, OpenAPI, or custom APIs).
Catalogs live on your domain, registries search across them
Organizations publish an ai-catalog.json file at a well-known path on their own domain. That domain ownership becomes the cryptographic anchor for identity and trust. Registries, acting like search engines for the agentic web, crawl these catalogs and index their capabilities. When a client agent needs something - say, a deployment history tool - it can query a registry with a plain-language intent, or fetch a known partner's catalog directly.
No central authority. No vendor lock-in. Any agent that speaks HTTP can participate.
Cryptographic verification before the handshake
Production deployments demand proof. ARD lets publishers attach verifiable trust metadata to their catalog entries. Before a client agent opens a connection, it can confirm the publisher's true cryptographic identity - not just a domain claim, but a signed assertion. This is ARD's trust manifest, and it's what enables enterprise compliance requirements like HIPAA to be met without sacrificing interoperability.
Once trust is established, ARD steps out of the way. The agent connects directly to the tool using its native protocol. No proxying, no middleware, no bottleneck.
Google's Agent Registry turns the spec into an enterprise product
Google Cloud is shipping ARD as a first-class feature in the Gemini Enterprise Agent Platform. Agent Registry provides fully hosted catalog hosting, federated search, and governance controls: globally unique URNs, egress policies, tool pinning, and Agent Identity for trust manifest verification. Native ARD support arrives in the coming months, letting organizations connect their internal registries to the broader federated network.
The ARD specification is available now, with a quickstart guide to publish your first ai-catalog.json in minutes. The agent web doesn't need another protocol - it needs one way to find the ones already out there. ARD might be that layer.
Source: Announcing the Agentic Resource Discovery specification
Domain: developers.googleblog.com
Comments load interactively on the live page.