Agent card
A machine-readable JSON file at a well-known URL describing what an agent can do and how to reach it — a business card other agents can parse.
01What it means
An agent card is a small JSON document at a predictable location — by convention /.well-known/agent-card.json — that describes an agent: its identity, what it can do, where its endpoint lives, and what authentication it expects. Before one A2A agent sends work to another, it reads the card to learn what that agent offers and how to address it. The MCP equivalent is the server card at /.well-known/mcp/server-card.json, describing a tool server the same way.
Cards do for the agent ecosystem what sitemaps did for the web: no central registry, no human paperwork — an agent that knows the convention can find the capability on its own. This site publishes both card types for its read-only site agents, which is how agent frameworks discover them without being told.
Because a card is a file you control, it doubles as a statement of scope: what the agent offers, nothing more. Ours, for instance, advertises read-only tools — search our content, list our services, estimate ROI — and nothing that writes.
02Why it matters for buyers
- —A card is an interface contract in miniature: capabilities, endpoint, and auth requirements stated in a parseable file rather than a sales conversation.
- —It is trivially checkable: our agent-readiness tool looks for both A2A agent cards and MCP server cards on any public site — presence is a readiness signal, scope is the substance.
- —Absence is informative too: most sites publish none yet, so publishing yours is a cheap differentiator with the agents — and the buyers using them.
03Common questions
Where does an agent card live?
At a well-known URL: /.well-known/agent-card.json for A2A agents, /.well-known/mcp/server-card.json for MCP servers. The predictability is the point — no discovery service required.
What is inside one?
Identity and description, capabilities or skills, the service endpoint, and authentication metadata — enough for another agent to decide whether to delegate and how to address the request.
Is an agent card the same as robots.txt?
No. robots.txt publishes policy for crawlers — what may be fetched. An agent card advertises an active capability another agent can call. One states rules; the other states an offer.