Agent Card

An Agent Card is a JSON document an AI agent publishes about itself, usually at a predictable, well-known URL, so anyone, human or another agent, can look it up before interacting with it and learn what it does, how to call it, and how to authenticate to it. The idea solves a specific coordination problem: as soon as agents start talking to other agents they didn't write, both sides need a shared way to answer "what is this thing, and how do I use it?" without a human reading documentation first.

The concrete shape of an Agent Card, in the specification that actually uses that name, includes fields like a name and description, a provider, a version, a capabilities block (does it support streaming responses? push notifications?), an authentication section (what credentials does a caller need?), and (the part that does the real discovery work) a skills array: a structured list of the specific tasks the agent can perform, each with its own description and example inputs. A caller reads the card once and knows whether the agent can do what it needs, without a trial-and-error API call.

Where it comes from

The AgentCard concept and its concrete JSON schema belong to Google's Agent2Agent (A2A) protocol, announced in April 2025 as a way for AI agents built by different vendors, on different frameworks, to discover and talk to each other. Google later contributed the protocol to the Linux Foundation to give it a vendor-neutral governance home, rather than keeping it a single-company standard. Solidus did not design the AgentCard schema, is not a contributor to the A2A specification, and does not implement the A2A protocol.

Solidus status

Solidus Agents serves a document it calls a "SolidusAgentCard" at /v1/agents/:id/card, the same URL that an agent's on-chain ERC-8004 passport (see Agent Passport) points to. It's A2A-style in spirit, a discoverable JSON document describing an agent, but it is not A2A-conformant: it doesn't carry the skills, capabilities, authentication, or input/output-mode fields the real A2A schema requires. What it carries instead is a did:solidus identity, that DID method's current registration status, an optional link to an ERC-8004 Identity Registry token, and the agent's active BBS+ credentials. It answers a different question than A2A's card does: not "what can this agent do," but "who issued this agent an identity, and what has been verified about it." Calling both artifacts "AgentCard" risks implying interoperability that doesn't exist, a caller who expects a real A2A skills array from Solidus's card will get nothing, because Solidus never populates one. This runs on Solidus's public testnet only.

Proof you can run yourself: curl https://agents.solidus.network/v1/agents/{id}/card returns the live SolidusAgentCard JSON. Compare its top-level keys against the published A2A AgentCard schema (a2a-protocol.org) to see the gap directly: Solidus's document and A2A's are structurally different artifacts that happen to share a naming convention.