precision-bio-tests

name	precision-bio-tests
description	Expert guide for testing bioinformatics MCP servers. Covers pytest, DRY_RUN modes, and PatientOne simulation scenarios.

Precision Bio-Testing

This skill provides guidelines for validating the complex biological and clinical logic in this repository.

Automated Testing Strategy

1. Unit Testing (pytest)

Tests live in two locations:

• Project-root tests: tests/unit/mcp-[server]/ (e.g., tests/unit/mcp-spatialtools/)
• Server-local tests: servers/mcp-[name]/tests/ (e.g., servers/mcp-perturbation/tests/)

Both patterns are in active use. Check both locations when looking for existing coverage.

2. The .fn Testing Pattern

FastMCP Pattern: Never call @mcp.tool() decorated functions directly. Always use the .fn attribute to bypass the MCP protocol layer:

result = await server.perturbation_predict.fn(gene="BRCA1", cell_type="T-cell")

This pattern is used across 9 test files in the project. Always follow it when writing new tests.

3. DRY_RUN Mode

Always test both modes:

• DRY_RUN=True (default) — returns synthetic/simulated data, no real bio tools needed
• DRY_RUN=False — exercises real logic (monkeypatch external dependencies in CI)

4. Bioinformatics Fixtures

Some test suites use fixture files (e.g., tests/unit/mcp-multiomics/fixtures/ contains sample CSV files). Large datasets should be mocked or referenced from data/patient-data/.

The PatientOne Scenario

The gold standard for integration testing is the PatientOne workflow (docs/reference/testing/patient-one/):

• Phase 1: Clinical (Epic)
• Phase 2: Multi-omics
• Phase 3: Spatial
• Phase 4: Imaging
• Phase 5: Final integrated recommendation

Manual Testing (Prompts)

For browser-based or chat-based testing:

• Use the templates in docs/reference/testing/quick-test-prompts.md
• Verify the "Clinical Interpretation" matches the expected biological signal (e.g., GZMB/IFNG upregulation in immunotherapy)

Testing Checklist

• Tool discovery via MCP protocol works
• .fn pattern used for all tool function calls in tests
• Both DRY_RUN=True and DRY_RUN=False paths covered
• Memory usage remains within 4Gi during heavy bioinformatics processing
• Error messages are informative for clinicians, not just developers

---

Use this skill when:

• Adding new tools to an MCP server
• Running regression tests on patient data pipelines
• Validating a new server's "production-ready" status