a2a-testing

A2A Testing

Before writing code

Fetch live docs:

1. Fetch https://a2a-protocol.org/latest/specification/ for protocol requirements to test against
2. Web-search site:github.com a2aproject A2A testing conformance for testing tools and guidance
3. Web-search site:github.com a2aproject a2a-samples test for test examples
4. Fetch SDK docs for test utilities, mock classes, and test helpers

Conceptual Architecture

Testing Levels

A2A implementations need testing at multiple levels:

1. Unit Tests

Test individual components in isolation:

• Agent handler logic (given input message, verify output)
• Task state management (verify state transitions)
• Message/Part construction and parsing
• Agent Card validation
• Error code generation

2. Integration Tests

Test the A2A server with a real HTTP stack:

• Send JSON-RPC requests, verify responses
• Test all supported methods (message/send, tasks/get, etc.)
• Verify Agent Card serving at /.well-known/agent-card.json
• Test authentication flows
• Test streaming with SSE client

3. Protocol Conformance Tests

Verify compliance with the A2A specification:

• All required methods are implemented
• Request/response schemas match the spec
• Error codes are correct
• State transitions follow the rules
• Agent Card schema is valid

4. End-to-End Tests

Test full multi-agent workflows:

• Client discovers agent via Agent Card
• Client sends task, agent processes, client receives result
• Multi-turn conversations complete successfully
• Error scenarios are handled gracefully
• Multiple agents collaborate on a complex task

Mock Agents

For testing clients, create mock A2A servers that:

• Return predefined responses
• Simulate different task states
• Simulate errors and failures
• Simulate slow responses and timeouts
• Support streaming with controlled events

For testing servers, create mock A2A clients that:

• Send various message types and parts
• Exercise multi-turn flows
• Send invalid requests to test error handling
• Simulate disconnections during streaming

What to Test

Server tests:

• Agent Card is valid and accessible
• All declared methods work correctly
• Task state transitions are correct
• Error codes are appropriate
• Streaming events are properly formatted
• Push notifications are delivered
• Authentication is enforced
• Concurrent requests are handled

Client tests:

• Agent Card discovery and parsing
• Request construction (correct JSON-RPC format)
• Response handling for all task states
• Error handling and retry logic
• Multi-turn conversation flow
• Streaming event processing
• Authentication credential handling

Testing Patterns

Stateful server tests: Use an in-memory task store, send a sequence of requests, verify the accumulated state.

Snapshot testing: Capture JSON-RPC request/response pairs and verify they match expected schemas.

Chaos testing: Introduce random failures, slow responses, and disconnections to test resilience.

Contract testing: Verify that client and server agree on the message schemas (both sides validate against the spec).

Best Practices

• Test all 9 task states, not just the happy path
• Test invalid state transitions (should return errors, not crash)
• Test with all Part types (TextPart, FilePart, DataPart)
• Test streaming with slow consumers and fast producers
• Test authentication with valid and invalid credentials
• Use deterministic agent logic in tests (not real LLMs) for reproducibility
• Validate all JSON-RPC responses against the schema
• Run tests in CI/CD pipelines
• Test Agent Card changes (schema validation)

Fetch any official conformance test suites or testing utilities from the A2A project before building custom test infrastructure.