Google, along with Microsoft, GitHub, Cisco, NVIDIA, Salesforce, Hugging Face, GoDaddy, Databricks, ServiceNow, and Snowflake, has published the Agentic Resource Discovery (ARD) specification, an open standard for publishing, discovering, and verifying AI tools, skills, and agents across the web. The problem ARD targets is upstream of what MCP solves. MCP standardizes how an agent calls a tool once it knows the tool exists. ARD addresses an earlier, largely unsolved step: how an agent finds the right tool in the first place when capabilities are scattered across organizations that don't share a registry. As InfoQ put it, ARD works alongside existing protocols like MCP and OpenAPI for execution, adding a discovery layer on top rather than replacing them. The mechanics are simple by design. An organization publishes a machine-readable ai-catalog.json manifest at a well-known path on its own domain. Registries then crawl and index these catalogs, and agents can query them using natural-language, intent-based search instead of relying on hardcoded integrations. The spec answers three runtime questions: where does the right capability live, which one should I use, and how do I verify it's safe to connect to. The launch partner list is notably broad for a spec this early — it's licensed under Apache 2.0 and builds on the Linux Foundation's AI Catalog data model, which suggests an attempt to avoid the single-vendor-lock-in problem that has dogged earlier agent registries. Google is backing it commercially through Agent Registry in its Gemini Enterprise Agent Platform, which handles namespaced URNs, egress policy enforcement, and tool pinning, and which ties into Agent Identity to verify a cryptographic trust manifest before an agent connects to a resource. Early adopters are already visible: GitHub has built an Agent Finder into Copilot, and Hugging Face has shipped a Discover Tool on top of ARD. What's worth watching is whether ARD actually gets adopted as a shared web-scale layer or ends up as another catalog format that only the launch partners implement consistently. The pitch — an open, domain-based manifest anyone can publish, crawled by federated registries — is structurally similar to how robots.txt or sitemap.xml became web-wide conventions rather than platform features. Whether agent discovery follows that path depends on whether enough independent publishers adopt the ai-catalog.json convention without a registry operator forcing it.