Agent Card - the A2A protocol's identity file that lets agents find your agent
Tests for a /.well-known/agent-card.json document describing your agent to other agents.
What this signal tests
We check for an Agent Card at /.well-known/agent-card.json - the discovery document defined by the Linux Foundation's A2A (Agent-to-Agent) protocol. It declares your agent's name, description, URL, skills, capabilities, supported transports (JSON-RPC, gRPC, REST), and security schemes so that other agents can find and call yours programmatically.
Why it matters for your visibility in AI
A2A is the proposed standard for agent-to-agent commerce - the layer above MCP that lets one company's AI agent discover and negotiate with another's. Where MCP is about tools, A2A is about agents talking to agents. The protocol is young (graduated to Linux Foundation governance in mid-2025) and adoption is still small, but the big AI labs and several enterprise vendors are aligning on it. If your business model involves any kind of programmable transaction - fulfilment, logistics, scheduling between organisations, B2B procurement - an Agent Card now means another company's agent can discover your agent without a custom integration. Early movers in directories like a2a-protocol.org get cited as reference implementations, which is free distribution to every developer building on the protocol.
Pass criteria at a glance
| Criterion | Passes when |
|---|---|
| Valid JSON schema match. |
How we test it
We send `GET /.well-known/agent-card.json` and check three things. First, the response status is 200. Second, the body parses as valid JSON. Third, it contains the required A2A fields: `name`, `description`, `url`, `skills`, and `capabilities`. We also confirm the declared `url` matches the host we fetched from. If any of those checks fail - or the file is absent entirely - the signal fails. Content type should be `application/json`.
Show technical detection method
GET /.well-known/agent-card.json; pass if 200 + valid JSON with required fields.
If your site fails: how to fix it
- Read the A2A specification at https://a2a-protocol.org/latest/specification/ to understand the field requirements and your declared skills surface.
- Generate the Agent Card with an A2A SDK (Python or TypeScript reference implementations exist) rather than hand-writing JSON - the schema is evolving.
- Serve the card at exactly /.well-known/agent-card.json with Content-Type: application/json, cacheable but not cached for hours.
- Declare your supported transports (most implementations start with JSON-RPC over HTTPS) and reference your MCP endpoint if you have one.
- Because A2A is still a proposed standard, have a developer who can read the spec implement this and revisit quarterly as the schema stabilises.
Quick facts
| Maturity | PROPOSED |
|---|---|
| Weight | low |
| Category | Emerging Standards |
Primary sources
Related signals
Frequently asked questions
Is A2A a real standard or just a proposal?
It is a proposed standard under Linux Foundation governance, not an IETF RFC or W3C recommendation. The specification is public and stable enough for production trials, but the field set may still change before a 1.0 release.
Which agents honour the Agent Card today?
Adoption is early - primarily reference implementations from Google, Salesforce, ServiceNow, and a handful of agent platforms. Coverage is far behind MCP. Publishing one now is forward-looking infrastructure, not something most consumer chatbots read yet.
Do I need an Agent Card if I already have an MCP server?
They solve different problems. MCP exposes your tools to an AI assistant. An Agent Card describes your whole agent (identity, skills, supported protocols) so other agents can decide whether to talk to you at all. In practice the two will coexist.
Is there an easier alternative?
Not yet at the same scope. /llms.txt and an MCP endpoint together cover most of the discoverability ground for the AI assistants people actually use today. Add an Agent Card when your roadmap genuinely includes agent-to-agent transactions.
Run your own scan
Run a free scan and see how your site grades across all 155 AI-readiness signals.