The Agentic Resource Discovery specification, or ARD, proposes a standard way for an AI client to ask a single question: what tool, Skill, MCP server, API or agent can help with this task? Today that wiring is mostly manual, with a person finding a capability, judging whether to trust it, and connecting it by hand.
The premise is that the hard part has moved. "The bottleneck is no longer invocation. It is discovery," the spec argues. A client cannot use a capability it does not know exists. ARD only handles finding the right resource and describing what it does, who provides it and where it lives; invocation still runs through that resource's own mechanism, whether MCP, an API or an agent framework. The authors name GitHub's Agent Finder and Hugging Face's Discover as services that could implement it.
MCP standardised how agents call tools. ARD is a bet that the next contest is how an agent finds the right one among thousands published by every vendor, team and company.