Compute evidence-aware polygenic risk scores from local VCF files via a pinned just-prs MCP server.
npx skills add https://github.com/ClawBio/ClawBio --skill just-prs-mcp
Please help me install this Agent Skill. Skill: just-prs-mcp Source: https://github.com/ClawBio/ClawBio/tree/main/skills/just-prs-mcp Command: npx skills add https://github.com/ClawBio/ClawBio --skill just-prs-mcp
Computing polygenic risk scores from a personal VCF or WGS file normally requires manually finding PGS Catalog models, normalizing data, running scores, checking variant coverage, and interpreting percentiles against the right ancestry panel. This Skill automates that entire workflow through a validated, version-pinned local just-prs MCP server while keeping the genome file on the user's machine.
Ask the agent to compute an evidence-aware PRS report for type 2 diabetes from your local VCF file using the EUR reference panel, or run the offline demo with "clawbio run just-prs --demo".
High-intent search queries and answers about installing and using this skill.
Provide a local single-sample VCF (or VCF.GZ/BGZ) plus one selector: a trait term, an EFO/MONDO trait ID, or a PGS Catalog ID. The bridge launches a pinned just-prs-mcp server over local stdio, scores the models, and writes a report with percentiles, coverage, and quality labels.
just-prs-mcp accepts already-called VCF or WGS files, while gwas-prs handles 23andMe or AncestryDNA text exports. If your input is raw FASTQ or BAM, run variant calling with wgs-prs first before using this skill.
No upload is performed by the bridge. The VCF stays on your machine and only its resolved local path is passed to a local stdio child process; live runs fetch only public PGS Catalog metadata, scoring files, and reference distributions, and the offline demo makes no network calls.
Absolute risk is computed only when the percentile is marked reliable and upstream prevalence or effect-size evidence exists. It is also withheld when a genome-build mismatch is detected, because the underlying score is not interpretable across assemblies.
When --superpopulation is omitted, percentiles default to the EUR 1000 Genomes reference panel and a warning is emitted. Pass AFR, AMR, EAS, EUR, SAS, or AUTO explicitly so percentiles match the appropriate reference ancestry.
A free-text trait term can match multiple PGS Catalog ontology records, and the bridge refuses to guess. Re-run with --trait-id using one of the listed EFO or MONDO identifiers to get a deterministic result.
openaiPlan and run germline WGS, WES, and panel variant-calling workflows from FASTQ, BAM, or CRAM.
ClawBioGenerates unified genomic profile reports from uploaded JSON data.
mims-harvardBuild and interpret polygenic risk scores from GWAS summary statistics and genotype data.
ClawBioGenerate personalised nutrition reports from 23andMe, AncestryDNA, or VCF genetic data.
