Testing Framework (A2TF)
A²TF (Agentic Asset Testing Framework) is a no-code testing framework for MCP Servers and Agent Cards, with a test-first mindset.
Overview
A²TF lets you design test cases, build suites, and execute them against mock or production endpoints — without writing code. It supports MCP protocol and A2A protocol validation.
Key modules:
- Dashboard — Metrics, tests by type, execution summary
- Design — Test cases, datasets, and scenarios per asset
- Execution — Build suites and run tests
- Test Runs — Filter and inspect run history
Relation to Test Lab
Test Lab is the in-app entry point for execution and exploratory sessions. A²TF provides the full design and execution workflow and may be available as a dedicated subdomain (e.g., a2tf.a2d-ai.com).
Dashboard
Environments are linked to a specific asset and carry environment-specific configuration (base URL, auth, headers). Suites group one or more tests and can be run against a chosen environment.
Mocking Service Environment
When you create an MCP Server, Agent Card, or REST API, a default Mocking Service environment is automatically created. This points to A²D’s built-in mock server so you can test immediately.
When assets are imported via Import/Export, the Mocking Service environment is also created automatically.
Test Runs Module
Filter and inspect past runs:
- Status, test name, linked asset
- Date and time
- Tools and data used
- Open reports for full details
A2A protocol version in test execution
Agent-card tests run against the A2A protocol, and A²TF sends the wire format that matches the card. When you run a suite (from the UI or via Platform MCP), the runner:
- Reads the card’s
protocol_version(0.3.0,1.0, ordual). - Builds the JSON-RPC envelope in that dialect —
message/sendwithrole: "user"for0.3.0,SendMessagewithrole: "ROLE_USER"plus theA2A-Version: 1.0header for1.0. - Unwraps the response before evaluating assertions, so a condition path like
artifacts.0.parts.0.textworks for both versions (0.3.0returns theTaskflat underresult;1.0nests it underresult.task).
Per-test version override
For a dual card that speaks both dialects, you can pin which one a test exercises. In the test detail panel, the A2A protocol version dropdown offers:
- Auto (from card) — derive from the card’s
protocol_version(the default). - A2A 0.3.0 / A2A 1.0 — force that dialect.
An explicit choice only applies to dual cards; a single-version card always sends its own version regardless of the selection. To cover both dialects of a dual card, create two tests — one pinned to each version.
The override is stored on the test config as protocolVersion and is honored identically by the UI suite runner and the Platform MCP run_test tool, since both share the same execution route.
Platform MCP Integration
AI agents can run suites and fetch results via Platform MCP:
run_suite— Execute a test suiterun_test— Execute a single test by IDget_run_results— Retrieve run results by ID
Next Steps
- Test Lab — Execution, Exploratory, Test Runs
- Platform MCP — Automate test execution
- Protocol Tester — Manual MCP testing
A²TF makes agent asset testing accessible without code.