A new open specification called Agentic Resource Discovery, or ARD, aims to help AI agents locate tools and resources across multiple clouds and platforms without needing to know every connection in advance. The specification was authored by Junjie Bu of Google, R.V. Guha of Microsoft, and Shaun Smith of Hugging Face, and released under the Apache 2.0 license. AWS featured the project in its August 31 Weekly Roundup, calling it "DNS, but for agents," though the company's involvement has been limited to providing feedback and exploring integration with its own Agent Registry.
The current v0.91 proposal, dated August 26, uses JSON-LD and a REST interface. It requires a POST /search endpoint that searches by task, while optional endpoints let clients browse what's available. Each discovery service can establish its own rules for what it returns and which sources it trusts. Engineers from Cisco, Databricks, GitHub, GoDaddy, Nvidia, Salesforce, ServiceNow, and Snowflake have contributed to shaping the project. The specification uses "agentic resource" to describe anything an AI client can connect to, from a Model Context Protocol server to other external capabilities.
According to the report, the Model Context Protocol has become a common way for AI applications to connect to external tools and data, but it assumes the client already knows which server it wants to use. That becomes a problem as companies spread their infrastructure across clouds, SaaS platforms, and internal systems. ARD addresses this by helping an agent find a resource before it tries to use it. AWS describes the approach as "describe once, discover everywhere," allowing companies to maintain their own catalogs and policies while routing searches to other ARD-compatible services.
The specification enables federation without forcing migration. An enterprise could keep internal resources private while searching approved external catalogs when needed. An ARD-compatible service keeps track of what's available, rather than requiring developers to set up every connection ahead of time. This matters because connecting an AI agent to a single tool is relatively straightforward, but things get more complicated once an organization has hundreds or thousands of resources spread across different platforms. The approach differs from DNS in a key way: while a domain name points to a specific location, an ARD search could return several options that all appear capable of doing the job.
ARD's governance is still being worked out, with board terms and membership among the details yet to be settled. The group has discussed eventually moving the project to a neutral organization such as the W3C or an AI foundation. AWS is already exploring how it could connect with Agent Registry and find resources outside its own catalog. The report notes there may not be much time to settle on a common approach, since connecting all these directories will only get harder once companies have built their own discovery systems. For organizations racing to deploy agentic workflows at scale, the window to converge on shared infrastructure may close faster than the standards process typically moves.