Google has introduced Agentic Resource Discovery, an open specification designed to help AI agents find, verify and connect to tools, skills and other agents across different platforms. The company says Microsoft is among the partners supporting the effort, which aims to reduce fragmentation in the fast-growing agent ecosystem.
The proposal addresses a practical challenge for agentic systems. As agents increasingly rely on external capabilities, they need a reliable way to answer three questions: where a capability is located, which capability is the right one to use, and how to confirm it can be trusted before connecting. Google says there is no shared standard today for handling that process across organizations.
ARD, short for Agentic Resource Discovery, is meant to fill that gap. The specification defines a common way to publish, find and verify AI capabilities on the web, regardless of the framework, protocol or vendor behind them. According to Google, the goal is to make tools and services easier to share securely across organizational boundaries.
The system is built around two main pieces: catalogs and registries. Under the model, organizations publish catalogs on their own domains that describe the resources they make available. Those domain names are intended to serve as the basis for identity and trust. Registries then crawl and index those catalogs, acting as searchable directories for agents that need a capability at runtime.
When an agent looks for a service, a registry can return matching options along with trust data that helps confirm the publisher’s identity before a connection is made. Google says the design allows an agent to discover a resource through search, or bypass search altogether and fetch a known partner’s catalog directly from its domain.
The company described several types of resources that can be listed in a catalog, including MCP servers, A2A agents, OpenAPI tools and nested catalogs. Once a resource is selected, the client agent connects using the tool’s native protocol or API.
Google also tied the specification to its enterprise offerings. The company said native support for ARD will come to the Gemini Enterprise Agent Platform in the coming months. Within that product, Google Cloud offers an Agent Registry intended to help organizations search, discover and host agentic resources. The registry is also positioned as a governance layer, with features such as unique namespaced URNs, policy enforcement and support for identity checks.
The company said the trust layer is designed to help enterprises meet compliance and security requirements, including HIPAA. It added that authenticated publisher onboarding is planned for the Agent Registry.
Google said ARD is being released as an open specification and is licensed under Apache 2.0. It is built on the AI Catalog data model and draws on work from the AI Catalog Working Group under the Linux Foundation, along with launch partners that helped shape the specification and reference implementations.
The company is now inviting developers and organizations to publish catalogs, review the specification and contribute to the project through its GitHub repository. Google framed the release as part of a broader effort to keep the agent ecosystem decentralized and open as adoption expands.