Google unveils Agentic Resource Discovery, an open standard for AI agent collaboration

Google has announced Agentic Resource Discovery (ARD), an open specification designed to solve a critical fragmentation problem in AI agent ecosystems: how agents find, trust, and connect to distributed tools and services across organizational boundaries.

Currently, tools and capabilities are locked into proprietary registries within specific platforms. ARD aims to change that by creating a standardized, decentralized network where agents can dynamically discover resources regardless of their underlying framework or provider.

How ARD Works#

The specification relies on two core primitives:

• Catalogs: Organizations publish an ai-catalog.json file under their own domain, describing available capabilities—whether MCP servers, OpenAPI tools, or other agents. Domain ownership serves as the cryptographic foundation for trust.
• Registries: Search engines for the agentic web. They crawl published catalogs, index their contents, and help agents find matching capabilities with verifiable trust metadata.

The system follows four phases: a provider publishes its catalog, an agent queries a registry or fetches directly from a known partner, cryptographic verification confirms the publisher's identity, and the agent establishes a direct connection using the resource's native protocol.

Enterprise Implementation#

Google is backing this with Agent Registry in Gemini Enterprise Agent Platform, a fully hosted enterprise offering that provides governance, secure resource management using Agent Identity, compliance support (HIPAA), and integration with internal registries into the broader federated network.

The specification is available now under Apache 2.0. Developers can publish their first catalog in minutes, review the full specification and reference implementations on GitHub, and contribute to the evolving standard alongside launch partners and the AI Catalog Working Group under the Linux Foundation.