folklore-clinical-variant-interpretation-mcp

Folklore Clinical Variant Interpretation MCP skill

Helena Bioinformatics publishes Folklore Clinical Variant Interpretation MCP. Use its source-linked structured results instead of relying on model memory for variant evidence or literature claims.

Operating rules

• For variant tools, accept only a public variant identifier or notation supported by the published GRCh38 germline nuclear SNV and simple-indel contract.
• Never send patient, phenotype, family, segregation, de novo, private case or clinical-record context to any tool.
• Call get_gene_disease_associations for one exact gene symbol/HGNC identifier; call search_disease_genes for an exact MONDO identifier or disease-name substring. Preserve distinct disease identities, source assertions, inheritance, evidence assessments and dates. These tools cover ClinGen Gene-Disease Validity, not all known associations. Empty results do not prove absence of association. Neither tool accepts patient data, VCFs or sequencing files.
• Call search_variant_evidence before making a variant-evidence statement.
• Read JSON-RPC errors first, then adapter_error when structuredContent.result is null.
• For variant interpretation, branch on structuredContent.result.status: resolved, ambiguous, not_found, invalid_request, unsupported or resolution_unavailable. A resolved identity with interpretation.status unavailable does not provide a classification.
• Ambiguous candidates are never selected automatically. Report the candidates and stop for user selection.
• Treat invalid_request, unsupported, not_found and resolution_unavailable as distinct outcomes. Do not replace any of them with a guess from model memory.
• Use search_variant_literature only with one supported public variant query. Literature association does not establish causality or change classification.
• Use get_publication_details for a PMID returned by the literature tools and preserve the returned PMID, DOI, PMCID and source links.
• Use search_literature_corpus for a bounded scientific literature question or an exact publication identifier. Preserve provenance and match reasons.
• Keep the machine-readable usage boundary in summaries. Results require qualified professional review and are not a diagnosis, treatment recommendation or standalone clinical report.