Google, Microsoft, and GitHub Collaborate on New AI Agent Discovery Specification

A coalition of eleven technology companies, including Google, Microsoft, and GitHub, has unveiled Agentic Resource Discovery (ARD), an open specification aimed at streamlining how AI agents locate and validate tools, skills, and other agents across the internet. The draft specification, released on June 17, addresses a growing challenge in AI development where agents currently require pre-wired connections to every tool or API they utilize.

The ARD specification is designed to shift this process from manual pre-wiring to a dynamic search step performed at runtime. This move is particularly significant for companies that develop and publish AI tools and agents, offering a standardized method for their capabilities to be discovered and integrated.

Key facts:

Aspect	Detail
Release Date	June 17, 2026
Primary Goal	Standardize AI agent discovery and verification of online tools and capabilities
Contributors	Google, Microsoft, GitHub, Hugging Face, Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, Snowflake
License	Apache 2.0

The Problem ARD Aims to Solve

Currently, AI agents operate in a fragmented ecosystem where each tool or service integration often requires a dedicated, pre-configured connection. As the number of available AI capabilities expands rapidly, this approach becomes increasingly unscalable. ARD seeks to resolve this “coordination problem” by providing a common framework for discovery, enabling agents to find and utilize new resources on demand.

How ARD Works

The ARD specification introduces two core components: catalogs and registries. An organization publishes a catalog, typically an `ai-catalog.json` file hosted on its domain at a well-known path. This catalog lists the tools, capabilities, or APIs offered by that organization. Registries then crawl these catalogs, index their contents, and respond to discovery requests from AI agents in a natural language format.

Crucially, the specification leverages domain ownership for verification, enhancing trust in the listed resources. For production environments, publishers can embed cryptographic trust metadata, allowing agents or registries to confirm the publisher’s identity before establishing connections. Once a capability is selected, ARD facilitates the handover, and the agent connects directly using the tool’s native protocol.

Early Implementations and Google’s Role

Several contributing companies have already released working tools based on the ARD specification. GitHub, for example, introduced an agent finder feature for Copilot, enabling it to discover matching skills and tools from a chosen registry. Hugging Face launched a Discover Tool to search ARD-compliant services, and Cisco integrated the spec with its AGNTCY Agent Directory, an open-source project under the Linux Foundation.

Google’s contribution to ARD is centered around its Agent Registry, a component of the Gemini Enterprise Agent Platform. This registry is designed to host and search agentic resources, as well as manage enterprise governance. Google has indicated that native ARD support will be integrated into the platform in the coming months, allowing organizations to connect internal registries to the broader ARD network. It’s important to note that ARD is a specification for callable capabilities and not a direct feature of Google Search.

Implications for Indian Tech and Startups

For the Indian technology and startup ecosystem, ARD presents an opportunity to standardize how locally developed AI tools and services can be discovered and integrated by a wider array of AI agents. Companies building AI solutions, APIs, or specialized agents can leverage ARD to make their offerings more accessible and discoverable globally. This could foster greater interoperability and accelerate the adoption of Indian AI innovations within international AI workflows.

While ARD primarily targets publishers of callable capabilities rather than traditional content sites, its development underscores a broader industry trend towards a more machine-readable web. As the “agentic web” evolves, understanding and potentially integrating with such specifications could become crucial for tech companies looking to participate in the next generation of AI-driven applications. The v0.9 draft nature of ARD means it is still evolving, and its ultimate reach will depend on the growth of the registry ecosystem and broader industry adoption.

Source: Search Engine Journal, https://www.searchenginejournal.com/google-microsoft-back-draft-ai-agent-discovery-spec/579894/