AI agents are getting better at using tools. The harder problem is finding them. A new open protocol called Agentic Resource Discovery (ARD) addresses that gap directly, proposing a standardized way for agents to locate available resources across the fragmented systems that most enterprises actually run.
The scenario ARD targets is specific and familiar to anyone who has deployed an agent in a real organization. An agent investigating a production incident may need to pull from engineering documentation, open a ticket in a support system, check deployment history, and query an observability platform. Those systems likely live in separate registries, maintained by separate teams, with no shared discovery layer. The agent either fails, gets manually wired to each source, or requires a human to bridge the gaps. ARD proposes to collapse all of that into a unified lookup.
The protocol operates on two tiers. At the first tier, organizations publish catalogs that describe their available capabilities: what tools exist, where they live, and how to access them safely. At the second tier, registries crawl and index those catalogs, functioning the way a search engine functions for web pages. An agent queries the registry, gets back relevant tool references, and proceeds. InfoWorld reported on the specification, citing coverage by Maxwell Cooter, with a publication date of June 19, 2026.
ARD is worth distinguishing from MCP, Anthropic’s protocol for tool calling. MCP standardizes the interface between an agent and a tool it already knows about: how to invoke it, how to parse the response, how to handle errors. ARD operates one layer earlier, standardizing how an agent discovers which tools exist in the first place. The two protocols are complementary, not competing. An agent might use ARD to find a deployment-history tool and then use MCP to call it.
The backer list is substantial: Google, Microsoft, Cisco, Nvidia, and Salesforce are among the organizations behind the effort. The ARD specification is publicly available and the project has a GitHub community open to catalog publishers. A heavyweight coalition of this shape gives ARD a credibility floor that purely academic protocol proposals lack.
It does not guarantee adoption. The history of enterprise interoperability standards is littered with well-backed specs that stalled at the reference implementation stage. What typically differentiates the ones that spread is not who announced them but whether one major platform makes them load-bearing: if a dominant agent runtime requires ARD-compliant catalogs, adoption follows. None of the backers have disclosed whether that kind of platform commitment is on the table.
The governance question is also open. The specification is on GitHub, but the project’s decision-making structure, who controls the spec, who has merge authority, and how conflicts between backers get resolved, has not been publicly detailed. That distinction matters for organizations considering whether to publish catalogs now or wait for the standard to stabilize.
Teams building enterprise agents that need to span internal tool silos should track ARD closely over the next two quarters: if any of the major backers embeds catalog discovery into their agent runtime as a first-class feature, early catalog publishers will have a meaningful head start.
Reported by InfoWorld (Maxwell Cooter), published June 19, 2026.
