Skip to content

test-9-e2e-seqera-connector.md

Latest commit

 

History

History
114 lines (83 loc) · 7.19 KB

test-9-e2e-seqera-connector.md

File metadata and controls

114 lines (83 loc) · 7.19 KB
Raw

PatientOne End-to-End Test — Seqera Connector + Genomic Analysis

Last Updated: 2026-02-23

Purpose: Focused E2E test that demonstrates Seqera nf-core pipeline discovery alongside custom MCP servers for genomic analysis and reporting. Uses 3 custom MCP servers (DRY_RUN) plus the Seqera connector (live nf-core queries).

Prerequisites:

  • Claude Desktop with custom servers configured (desktop-configs/)
  • A free Seqera Platform account (cloud.seqera.io — free Cloud Basic tier, no credit card)

IMPORTANT — Connect to Seqera before running this test: Open Claude Desktop > Settings > Connectors > find Seqera > click Connect and complete the Seqera sign-in flow. The connector must show as authenticated (not just toggled on) before the Seqera tools will be available. See Verify Seqera Connector below.

See also: Base E2E test (no connectors) | E2E + Connectors (test-8) | Connector setup guide

Verify Seqera Connector Before Running

The Seqera connector connects to Seqera's infrastructure at mcp.seqera.io (unlike PubMed/ClinicalTrials/bioRxiv which are fully Anthropic-hosted). If authentication is incomplete, the connector will show as "connected" in the UI but expose zero tools — causing a Failed to fetch https://mcp.seqera.io/mcp error.

Quick verification — paste this into a new Claude Desktop conversation first:

What nf-core pipelines are available for somatic variant calling?

If Seqera tools are working, Claude will call nfcore_suggest_analysis and return pipeline recommendations. If you see "Failed to fetch" or a web search fallback instead, revisit Settings > Connectors > Seqera and re-authenticate.

Prompt

Copy and paste the following into Claude Desktop:

Run a PatientOne (PAT001-OVC-2025) precision oncology analysis focusing on pipeline selection and genomic interpretation. Uses 3 custom MCP servers (DRY_RUN) plus the Seqera connector for nf-core pipeline discovery (live).

Patient profile for context: 58-year-old female, Stage IV High-Grade Serous Ovarian Carcinoma (HGSOC), platinum-resistant, BRCA1 germline mutation, HRD-positive.

## Stage 1 — Clinical Context (mockepic)
Retrieve patient clinical data using mockepic:
- Patient demographics and diagnosis
- Lab observations (CA-125 trend, BRCA status)
- Current medications

## Stage 2 — Pipeline Selection (Seqera connector)
Use the Seqera connector tools (not web search) to recommend and explore nf-core pipelines for this patient's analysis. Call each tool directly:
1. Call the Seqera `nfcore_suggest_analysis` tool — suggest an appropriate nf-core pipeline for somatic variant calling in a BRCA1-mutated HGSOC tumor sample (WES data, GRCh38 reference)
2. Call the Seqera `describe_nfcore_module` tool twice — first for the "mutect2" module (somatic SNV/indel calling), then for the "cnvkit" module (copy number analysis)
3. Call the Seqera `search_nfcore_module` tool — search for modules related to "HRD" or "homologous recombination deficiency"
4. Summarize: recommended pipeline, key modules, and why they are appropriate for this patient

## Stage 3 — Somatic Variants & HRD (genomic-results)
Parse genomic results (simulating output from the recommended pipeline):
- parse_somatic_variants from "/data/patient-data/PAT001-OVC-2025/genomics/somatic_variants.vcf"
- parse_cnv_calls from "/data/patient-data/PAT001-OVC-2025/genomics/copy_number_results.cns"
- calculate_hr_deficiency_score using both files
- Summarize actionable mutations and PARP eligibility

## Stage 4 — Patient Report (patient-report)
First call get_report_template_schema to get the exact JSON schema, then construct valid JSON and call generate_patient_report:
1. Call get_report_template_schema — use the returned schema and example to build report_data_json
2. Build report_data_json as a JSON string with all 5 required sections: patient_info, diagnosis_summary, genomic_findings (list), treatment_options (list), monitoring_plan
3. Include findings from ALL prior stages: clinical context, pipeline recommendation, genomic findings
4. In treatment_options, reference the pipeline used and genomic evidence supporting PARP inhibitor eligibility
5. Call generate_patient_report with the JSON string, report_type="full", output_format="pdf"

## IMPORTANT — Final Output
After generating the report, display a final summary that includes:
1. A banner: "⚠️ DRAFT — REQUIRES CLINICIAN REVIEW AND APPROVAL BEFORE PATIENT DELIVERY"
2. Key findings from each stage (1-2 bullets each)
3. A "Pipeline Recommendation" section listing:
   - Recommended nf-core pipeline and why
   - Key modules (Mutect2, CNVkit, HRD-related)
   - Any relevant modules found for BRCA/HRD analysis
4. The report file_path from the generate_patient_report response
5. Instructions: "Use patient-report:approve_patient_report with reviewer_name to finalize"
6. Note which results are DRY_RUN synthetic data (Stages 1, 3, 4) vs live data (Stage 2)

Expected Results

Stage Server/Connector Data Source Key Expected Output
1 mockepic DRY_RUN Patient demographics, CA-125 trend, BRCA status, medications
2 Seqera connector Live nf-core/sarek recommended, Mutect2 + CNVkit module details, HRD module search results
3 genomic-results DRY_RUN TP53/PIK3CA/PTEN mutations, HRD score 44, PARP eligible
4 patient-report DRY_RUN Draft PDF path, validation passed

Final banner should read:

⚠️ DRAFT — REQUIRES CLINICIAN REVIEW AND APPROVAL BEFORE PATIENT DELIVERY

Notes

  • Custom servers (Stages 1, 3, 4) run in DRY_RUN mode — results are synthetic, not for clinical use
  • Seqera connector (Stage 2) queries the live nf-core module/pipeline registry — requires a free Seqera account for authentication (free Cloud Basic tier)
  • The 3 Seqera tools used (nfcore_suggest_analysis, describe_nfcore_module, search_nfcore_module) access the public nf-core registry only — they do not launch pipelines or incur compute costs
  • Typical runtime: 3-5 minutes (Seqera nf-core queries add ~1 minute vs the base test)
  • The report in Stage 4 is DRY_RUN (no real PDF generated) but the pipeline recommendation it references is grounded in real nf-core data

Troubleshooting

Symptom Cause Fix
Failed to fetch https://mcp.seqera.io/mcp Connector not authenticated Settings > Connectors > Seqera > disconnect and reconnect, complete sign-in
Seqera listed as connected but tools fall back to web search Auth token expired or tools not exposed Disconnect, reconnect, and re-authenticate
Stage 2 skipped entirely Connector not enabled Settings > Connectors > find Seqera > click Connect

See also: Base E2E test (no connectors) | E2E + Connectors (test-8) | Connector setup guide