Will Agent Resource Registries ever have their moment, the way AI gateways did?

ARD is an open discovery protocol for agentic resources (GitHub). It allows an AI client to ask: "What is available for this task?" and lets a discovery service answer with matching resources.

ARD sits entirely before invocation. It helps the client find the right resource; the resource is then invoked through its own native mechanism.

An agentic resource is any external capability an AI client can call on to do a task — an agent, MCP server, Skill, Canvas, Plugin, API, or workflow — anything that can be described as an ARD entry.

It is not an execution runtime: ARD is not MCP, A2A, Skills, AI Catalog, or an API runtime, and it does not replace them.

It is not a central catalog: There will be many discovery services, each indexing different resources, serving different communities, and applying its own trust, ranking, and access policies. Every enterprise can run its own — much like an intranet has its own search engine over internal content. And on the public web there will be many, some optimizing for quality and trust, others for coverage.

ARD is being developed by a working group with participants from Microsoft, Google, Hugging Face, GoDaddy, and others. This work is part of a broader effort to create an open discovery layer for resources that AIs can draw on.

To understand the motivation and design, start with the Introduction.

Agentic Resource Discovery (ARD) and its resources are licensed under the Apache License 2.0 and welcomes contributions from the community.