Automated Testing

SK Wwise MCP has three layers of testing — unit, integration, and eval — all built on Python and pytest.

Unit Tests

~450 tests across 27 files in tests/unit/. All WAAPI calls are mocked, so no Wwise installation is required.

Coverage includes:

• Core modules — WAAPI dispatcher, query engine, audio conversion, CLI wrapper
• Business logic — pipeline operations, transport, profiling, object manipulation
• MCP servers — one test file per server (browse, objects, containers, pipeline, etc.)

uv run pytest tests/unit/ -v

Integration Tests

8 test files in tests/integration/, marked with @pytest.mark.integration. These run against a live Wwise headless server with a test project.

What they test:

• Object queries, creation, deletion, and modification
• Switch/Blend Container assignments
• Audio import and SoundBank operations
• Transport/playback operations
• Audio peaks and Media Pool reads
• Generic WAAPI passthrough

The conftest.py fixtures handle setup automatically:

1. Locate WwiseConsole via WWISEROOT or the Windows registry
2. Start a headless WAAPI server on port 8080
3. Wait up to 20 seconds for the connection
4. Clean up leftover test objects before and after each run

uv run pytest tests/integration/ -v

To skip integration tests when running the full suite:

uv run pytest tests/ -v -m "not integration"

Eval Tests (LLM Routing)

39 test cases in tests/eval/ that verify the AI agent selects the correct MCP tools for a given prompt. This catches routing regressions when tools or server configurations change.

Each case defines a prompt and the expected tool(s):

{
  "id": 1,
  "prompt": "Ping Wwise to check if it's available",
  "expected_tools": ["ping_wwise"],
  "category": "browse"
}

Categories: browse, audition, generic, objects, media-read, cross-server.

Running Evals

Eval runs use Claude Code skills for orchestration:

1. Setup — create a test project and start the WAAPI server:

   /eval-setup

2. Run cases — choose batch (5 cases/invocation) or granular (1 case):

   /loop 60s /eval-batch
   /loop 30s /eval-routing

3. Generate report:

   python tests/eval/report.py

4. Cleanup:

   /eval-teardown

A PostToolUse hook in .claude/settings.json automatically logs every Wwise MCP tool call to tests/eval/tool_log.jsonl during eval runs. The verify.py script compares actual vs. expected tools and writes results to test_results.json.

Artifacts

Prerequisites

Install dev dependencies:

uv sync   # installs runtime + dev deps, creates .venv