DRY_RUN mode enables development and testing with realistic synthetic responses, without requiring:
- External tool dependencies (fgbio, STAR, DeepCell, etc.)
- Real data files or GCS access
- API credentials or authentication
- Large compute resources
Set the server's DRY_RUN environment variable to true:
export {SERVER_NAME}_DRY_RUN=true| Server | Variable |
|---|---|
| mcp-spatialtools | SPATIAL_DRY_RUN |
| mcp-fgbio | FGBIO_DRY_RUN |
| mcp-multiomics | MULTIOMICS_DRY_RUN |
| mcp-perturbation | PERTURBATION_DRY_RUN |
| mcp-deepcell | DEEPCELL_DRY_RUN |
| mcp-cell-classify | CELL_CLASSIFY_DRY_RUN |
| mcp-openimagedata | IMAGE_DRY_RUN |
| mcp-mocktcga | MOCKTCGA_DRY_RUN |
| mcp-genomic-results | GENOMIC_RESULTS_DRY_RUN |
| mcp-quantum-celltype-fidelity | QUANTUM_DRY_RUN |
| mcp-deidentify | DEIDENTIFY_DRY_RUN |
- Local development — Test without installing bioinformatics tools
- CI/CD pipelines — Automated testing with predictable outputs
- Demonstrations — Show capabilities without real patient data
- Onboarding — New developers can explore immediately
- Education — Classroom use with ~$0.32 total cost
Each server returns clinically realistic synthetic data based on the PatientOne case study. Responses include:
- Realistic data structures matching production output
- Biologically plausible values calibrated to published studies
- Appropriate metadata and annotations
- DRY_RUN: ~$1 total for full PatientOne workflow (25-35 minutes, tokens only)
- Production: Low per-patient compute cost (Cost Analysis)
Always set DRY_RUN=false in production deployments. Production mode requires:
- Real data files accessible via GCS or local paths
- Bioinformatics tools installed (or Cloud Run containers)
- Appropriate API credentials
mcp-deidentify note: When
DEIDENTIFY_DRY_RUN=false, the server callsclaude-haiku-4-5-20251001via the Anthropic API.ANTHROPIC_API_KEYmust be set. In Cloud Run (HOSPITAL1), this is provided via ambient service account auth — no Secret Manager action required.
Server installation: server-installation.md Cost analysis: cost-analysis.md