just-prs-mcp
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@just-prs-mcpsearch for type 2 diabetes scores"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
just-prs-mcp
An MCP server that wraps just-prs — a Polars-based tool for Polygenic Risk Scores (PRS) from the PGS Catalog. It exposes catalog search, PRS computation, VCF/array normalization, percentile & absolute-risk estimation, quality assessment, and (in extended mode) bulk downloads and reference-panel / pgen scoring as MCP tools.
Built on uv + FastMCP.
Agents: start with AGENTS.md.
Highlights
Hybrid tool registration with modes — a small always-on
essentialssurface (catalog lookup + the core compute/analyze workflow) plus anextendedsurface (batch, bulk downloads, HuggingFace upload, reference/pgen scoring) registered on opt-in. Keeps the default tool list small.Real background tasks — slow operations (
normalize_vcf,compute_prs_batch, downloads, reference scoring) run as@mcp.tool(task=True)with FastMCP's in-memory backend (no Redis);FASTMCP_DOCKET_URLswitches to Redis for scale.Structured I/O via Pydantic models + tool annotations — just-prs's own
PRSResult/AbsoluteRisk/TraitInfoare returned directly where they fit.Boot-safe — the server starts with no environment configured; no API key is required for any core feature.
In-memory test harness (
Client(transport=server)) — fast, deterministic, network-free.
Related MCP server: skore-mcp
Quickstart
uv sync # install deps (incl. dev)
uv sync --extra reference # + pgenlib for reference/pgen tools (Linux/WSL)
uv run pytest # tests, all in-memory (no network)
uv run just-prs-mcp stdio # run over stdio
uv run just-prs-mcp stdio --mode extended # expose the full tool surface
uv run just-prs-mcp http # run over HTTP (default :3011)
uv run fastmcp dev fastmcp.json # MCP InspectorThe server boots with no environment configured.
Tools
Tool | Tier | Notes |
| essentials | Search the PGS Catalog by free text |
| essentials | Cleaned metadata for one PGS ID |
| essentials | Best evaluation metrics (OR/HR/AUROC/C-index) |
| essentials | REST trait search |
| essentials | Trait by EFO ID + associated PGS IDs |
| essentials | VCF → genotype Parquet (background task) |
| essentials | Score one VCF against one PGS |
| essentials | Percentile of a PRS value (reference/theoretical/AUROC) |
| essentials | Absolute disease risk from a PRS z-score |
| essentials | Quality label + interpretation (pure logic) |
| extended | Score one VCF against many PGS (background task) |
| extended | 23andMe / AncestryDNA → Parquet (background task) |
| extended | One harmonized scoring file from EBI FTP |
| extended | All PGS IDs on EBI FTP |
| extended | All metadata sheets as Parquet (background task) |
| extended | Many/all scoring files (background task) |
| extended | Upload cleaned catalog to HuggingFace (needs token) |
| extended | Fetch 1000G / HGDP+1kGP panel (background task) |
| extended | Score against a panel (needs |
| extended | PLINK2 binary ops (needs |
Plus a resource (resource://prs/panels) and a prompt (compute_prs_for_trait).
File paths: computation tools take local paths (VCF / normalized Parquet /
.pgendir) on the server's filesystem. Over stdio that's your machine. Reference / pgen tools need the optional nativepgenlib(Linux/WSL —uv sync --extra reference); without it they return a clear install hint.
Modes
PRS_MCP_MODE (env) or --mode (CLI), default essentials:
essentials— catalog lookup + the core compute/analyze workflow.extended— everything (batch, bulk downloads, HF upload, reference/pgen).
Configuration
All PRS_MCP_* env vars are optional (see .env.example and settings.py):
PRS_MCP_MODE, PRS_MCP_CACHE_DIR, PRS_MCP_DEFAULT_GENOME_BUILD,
PRS_MCP_DEFAULT_PANEL, PRS_MCP_DUCKDB_MEMORY_LIMIT, PRS_MCP_HF_TOKEN,
PRS_MCP_TRANSPORT, PRS_MCP_HOST, PRS_MCP_PORT, PRS_MCP_LOG_LEVEL.
PRS_MCP_CACHE_DIR sets the root for cached catalog metadata, scoring files, and
reference panels; if unset, just-prs uses its own default (PRS_CACHE_DIR /
platformdirs). The HuggingFace upload tool reads PRS_MCP_HF_TOKEN or the native
HF_TOKEN.
Using with coding agents
.mcp.json (Claude Code) launches uv run just-prs-mcp stdio. For Codex
(~/.codex/config.toml):
[mcp_servers.just-prs]
command = "uv"
args = ["run", "just-prs-mcp", "stdio"]Deployment
Docker:
docker build -t just-prs-mcp . && docker run -p 3011:3011 just-prs-mcp(defaults to HTTP).Smithery:
uv sync --extra smithery; entrypoint inpyproject.toml[tool.smithery]+smithery.yaml.Declarative:
fastmcp.jsondrivesfastmcp run/fastmcp dev.
Project layout
src/just_prs_mcp/
server.py build_server(), CLI, graceful shutdown, Smithery entrypoint
settings.py pydantic-settings (PRS_MCP_*), safe defaults
client.py shared PRSCatalog / REST-client construction + adapters
models.py Pydantic tool I/O models (+ reused just-prs models)
logging_setup.py stdlib logging -> stderr
tools/
catalog.py essentials — PGS Catalog lookup
compute.py essentials — normalize / compute / analyze
extended.py extended — batch, downloads, HF upload
reference.py extended — reference-panel / pgen scoring (pgenlib)
tests/ in-memory client tests (wiring, not just-prs correctness)License
MIT — see LICENSE.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/winternewt/just-prs-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server