Eleven companies, including Google, Microsoft, GitHub, and Hugging Face, have published Agentic Resource Discovery (ARD). The open specification defines how AI agents discover and authenticate tools, capabilities, and other agents across the web.
Contributors released a draft spec on June 17, along with reference implementations for many of them. ARD is licensed under Apache 2.0 and builds on the AI Catalog data model maintained by a working group under the Linux Foundation. The full list of contributors includes Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, and Snowflake.
What Does ARD Solve?
The spec plans to solve the linking problem. Today, an agent must be connected to each tool, MCP server, or API that uses it ahead of time. As more companies publish their capabilities, that previous wiring stops measuring.
ARD moves detection to a search step that occurs at runtime. The change mainly affects companies that publish tools and agents, not traditional content sites yet.
How does ARD work?
ARD relies on two components, a spec that calls for catalogs and registers.
The organization publishes a catalog, i ai-catalog.json a file hosted in a well-known format in its domain, which lists the tools, MCP servers, agents, or APIs it makes available.
Registries then clear those catalogs, index, and respond to discovery requests from agents in plain language.
Because each catalog resides in its own publisher’s domain, the spec uses domain ownership to verify who published it.
For production use, publishers can attach trusted metadata so that an agent or registrar can verify the publisher’s cryptographic identity before connecting. Once the capability is selected, ARD closes and the agent connects directly using the device protocol.
Same Day Processing
Several contributors submitted work tools built on the spec on the same day.
GitHub has introduced an agent finder, which allows Copilot to find similar MCP servers, capabilities, tools, and agents from a selected registry, with users controlling that connection.
Hugging Face has released a Discovery Tool that searches MCP skills and servers across ARD services. Cisco has tied the spec to the AGNTCY Agent Directory, an open source project under the Linux Foundation.
The release advances the sequence of open data aimed at the machine-readable layer of the web. Google published the Open Knowledge Format, a specification for sharing organizational knowledge between AI systems, two days ago. The pattern in all these efforts is the same. Each one asks you to publish a structured file under your domain so that AI systems can use your content, without anyone having to manually wire it up.
Where Google Fits
Google’s role is focused on Agent Enrollment, which is part of its Gemini Enterprise Agent Platform.
The company said the Agent Registry will host and search agency services and handle corporate governance. Local native ARD support is planned for the coming months, which Google said will allow organizations to connect internal registrations to a wider network. That support isn’t live yet, and ARD is a feature instead of a Google search feature.
Why This Matters
Classification depends on what you publish. ARD is for publishers of callable capabilities, APIs, MCP servers, and agents that the software connects to. A publishing company has a clear path to being discovered and trusted by agents. A typical content site has no clear action to take today.
The importance 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-centric strategies. ARD targets tools and agents, not content, to raise questions about building now on systems that may or may not generate traffic in the long run.
Looking Forward
The spec is a v0.9 draft, and contributors invite changes via the GitHub project site. Its reach depends on registrations that can be transparent and index catalogs at scale, and that ecosystem is in its early stages. Support for Google Agent Registry has been out for months.
If that network develops, the profits mostly go to companies that provide the tools and agents that others need. The early features of the agent web from Google suggest this. The concern then is whether your current platforms and tools will support ARD and what they will require you to publish.
