Discovery
Publish an Agent Card, commonly at /.well-known/agent-card.json, with skills, endpoints, protocol version, and security schemes.
Stable v1.0.1 · Verified May 28, 2026
Build and evaluate interoperable agent workflows
A2A v1.0.1 defines how agents publish capabilities, exchange messages, manage tasks, and return artifacts across JSON-RPC, gRPC, or HTTP+JSON. A useful implementation still needs clear approvals, security, status, and failure handling.
Execution model
Use messages for communication, tasks for stateful work, and artifacts for outputs that clients can inspect or download.
Compatibility
Declare and negotiate versions and bindings. Use the maintained conformance kit and Inspector instead of assuming two implementations interoperate.
Published integrations
Directory inclusion describes a submitted integration; it is not a security audit or protocol certification. Verify capabilities and permissions with the maintainer.
Review the contract, not just the demo
A polished happy path says little about how an integration handles sensitive data, partial failures, or destructive actions.
Inputs and triggers
Accepted payloads, authentication, rate limits, and the event that starts work.
Actions and side effects
Systems changed, data written, messages sent, and operations that require approval.
Status and errors
Progress states, retries, idempotency, timeouts, cancellation, and actionable failure messages.
Ownership
Maintainer, support channel, release policy, data retention, and security disclosure route.