Compare

ARD vs A2A

A2A carries the conversation. ARD decides who you should be having it with.

A2A is how one agent talks to another. ARD is how it finds which agent to talk to. A2A's own discovery works when you already know an agent's domain; ARD indexes agents across publishers who have never coordinated, and returns the Agent Card's location so A2A can take over. In practice A2A agents are still a small share of published resources: 79 in our index against 8,714 MCP servers.

Two layers, often confused

A2A gives an agent a way to talk to another agent, and an Agent Card describing what it can do and how to reach it. That card is discoverable if you already know the agent's domain.

ARD answers the question before that one: out of every agent published anywhere, which one should I be talking to? It indexes A2A agents alongside MCP servers, skills and APIs, and returns the Agent Card's location as the entry's url.

	ARD	A2A
Answers	Which agent, out of all of them	How do I talk to this agent
Artifact	An entry in a manifest	An Agent Card
Discovery model	Search across publishers who never coordinated	Fetch from a domain you already know
Cross-vendor index	Yes, that is the point	No, per-agent
Covers non-agents	Yes: MCP servers, skills, APIs, registries	No, agents only

An A2A agent as an ARD entry

the url points at the Agent Card, not at a web page

{
  "identifier": "urn:air:acme.com:agent:support",
  "displayName": "Acme Support Agent",
  "type": "application/a2a-agent-card+json",
  "url": "https://api.acme.com/.well-known/agent-card.json",
  "description": "Opens, updates and escalates Acme support tickets.",
  "representativeQueries": [
    "open a support ticket with Acme",
    "check the status of my Acme ticket",
    "escalate an unresolved Acme issue"
  ]
}

Note what the url holds. The specification is careful here: the entry's url addresses the artifact, the Agent Card itself, and not the ARD entry that describes it and not a page about the product. Pointing it at a marketing page is valid and useless.

A2A agents are a small share of what is actually published. In our index there are 79 of them against 8,714 MCP servers. The agent-to-agent web is real but early, and most capability today is tools rather than agents.

Where they overlap, and where they do not

Both have a concept of describing yourself at a well-known path. That is where the similarity stops. A2A never tries to index anyone else, and ARD never tries to carry a conversation. The specification's FAQ makes a similar point about the Agent Name Service: find the resource with ARD, trust the name behind it with ANS, talk to it with its own protocol.

Questions

Is A2A part of ARD?

No. They are separate specifications with separate authors. ARD treats an A2A agent as one artifact type among several and identifies it with the media type application/a2a-agent-card+json.

A2A has its own discovery. Why do I need ARD?

A2A's discovery answers what a specific agent can do, once you know which agent to ask. It does not answer which of ten thousand agents you should be asking. That question needs an index, and an index is what ARD standardises.

Which should I publish first?

An Agent Card, because it is what an A2A client needs to actually talk to you. Then an ARD entry pointing at it, because that is what makes anyone find the card.

Last reviewed 2026-09-04. Checked against ARD v0.91 (Proposal, 2026-08-26).