Eleven companies, including Google, Microsoft, GitHub, and Hugging Face, have released Agentic Resource Discovery (ARD). THE open specification explains how AI agents find and verify tools, skills, and other agents on the web.
Contributors released the draft specification on June 17, along with reference implementations of several of them. ARD is licensed under Apache 2.0 and builds on the AI Catalog data model maintained by a Linux Foundation working group. The full list of contributors also includes Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow and Snowflake.
What does ARD solve?
The specification aims to solve a coordination problem. Today, an agent must be logged in in advance to every tool, MCP server, or API it uses. As more companies release their own capabilities, this pre-wiring stops evolving.
ARD moves discovery to a search step that occurs at runtime. This change primarily affects companies that publish tools and agents, not traditional content sites at this time.
How ARD works
ARD relies on two elements, which the specification calls catalogs and registers.
An organization publishes a catalog, a ai-catalog.json file hosted on a well-known path on its own domain, which lists the tools, MCP servers, agents, or APIs it makes available.
Registries then explore these catalogs, index them, and respond to agents’ discovery requests in plain language.
Since each catalog is on its own publisher’s domain, the specification uses domain ownership to check who published it.
For production use, publishers can attach trusted metadata so that an agent or registry can confirm the cryptographic identity of the publisher before connecting. Once a feature is selected, ARD makes the transfer and the agent connects directly using the tool’s own protocol.
Same day implementations
Several contributors shipped working tools built on the specification on the same day.
GitHub introduced the Agent Finder, which allows Copilot to discover matching MCP servers, skills, tools and agents from a chosen registry, with users controlling what is connected.
Cuddly face has released a discovery tool that searches for skills and MCP servers in ARD services. Cisco linked the specification to its AGNTCY Agent Directory, an open source project under the Linux Foundation.
This release continues a series of open specifications aimed at the machine-readable layer of the Web. Google published the Open knowledge formata specification for sharing organizational knowledge between AI systems, two days earlier. The pattern of these efforts is the same. Each requires you to post a structured file under your own domain so that AI systems can use what you expose, without anyone hand-wiring the connection.
Google’s place
Google’s role centers on Agent Registry, part of its Gemini Enterprise agent platform.
The company said Agent Registry will host and search agent resources and manage corporate governance. Native support for ARD in the platform is planned for the coming months, which Google says would allow organizations to connect internal registries to the broader network. This support is not yet available and ARD is a specification rather than a Google Search feature.
Why it matters
The distribution depends on what you post. ARD is intended for publishers of callable features, APIs, MCP servers and agents to which software connects. A company that publishes tools has a clear method for being found and approved by agents. A typical content site has no clear actions to take today.
The value of this effort is debated. Google’s John Mueller argued that LLM systems cannot use files like llms.txt to distinguish one site from another, and advised focusing on current needs rather than future agent-oriented strategies. ARD targets tools and agents, not content, which raises questions about building systems now that may or may not generate traffic later.
Looking to the future
The specification is a draft of version 0.9 and contributors are inviting changes via the project’s GitHub repository. Its reach depends on registries capable of crawling and indexing catalogs at scale, and this ecosystem is still in its infancy. Support for Google’s agent registry is months away.
If this network grows, the advantage primarily goes to companies offering tools and agents that others need. The beginning Google Web Agent Features allude to this. The immediate concern is whether your current platforms and tools will adopt ARD and what they will ask you to publish.
