Mendelian Randomization (ToolUniverse Claude Skill)

A ToolUniverse agent skill that answers “does X actually cause Y, or is the association confounded?” using pre-computed and custom two-sample Mendelian randomization over IEU OpenGWAS and EpiGraphDB MR-EvE.


Type	Claude Skill (one of ToolUniverse’s pre-built agent skills)
Supplier	Zitnik Lab, Harvard Medical School
Availability	GA — part of the ToolUniverse skills collection (`skills/tooluniverse-mendelian-randomization/`)
Pricing	Free / OSS (Apache-2.0); IEU OpenGWAS and EpiGraphDB are free academic APIs
Capabilities	Read-only — drives ToolUniverse tool calls; no data writes

How to install

This skill calls ToolUniverse tools, so the ToolUniverse MCP server must be installed first (see the ToolUniverse page). Simplest registration:

claude mcp add --transport stdio tooluniverse -- uvx tooluniverse

Then add the skills:

Claude Code — install the whole skill collection (the skill resolves as tooluniverse-mendelian-randomization):
```
npx skills add mims-harvard/ToolUniverse
```
Manual / other agents — copy just this skill directory into your skills folder:
```
git clone https://github.com/mims-harvard/ToolUniverse
cp -r ToolUniverse/skills/tooluniverse-mendelian-randomization ~/.claude/skills/
```
(replace ~/.claude/skills/ with your agent’s skills directory if you are not using Claude Code/Desktop.)

The skill sets disable-model-invocation: true upstream, so invoke it explicitly (e.g. ask Claude to “use the Mendelian randomization skill”) rather than relying on automatic dispatch.

What it does

Runs a five-step causal-inference workflow over four ToolUniverse tools:

Trait resolution — EpiGraphDB_search_opengwas maps the exposure and outcome you named onto exact OpenGWAS study labels, which is where most MR queries silently go wrong.
MR execution — EpiGraphDB_get_mendelian_randomization returns pre-computed exposure → outcome estimates from the MR-EvE graph; OpenGWAS_get_mr_instruments assembles instruments for a custom two-sample analysis when the pre-computed pair is missing.
Interpretation — effect direction and magnitude are read alongside instrument quality via the MOE score, so a large effect from weak instruments is not reported as a finding.
Triangulation — method agreement across IVW, MR-Egger and weighted median; bidirectional MR to rule out reverse causation; and EpiGraphDB_get_genetic_correlations to distinguish shared heritability from causation.
Drug-target follow-up — an optional step carrying a supported causal exposure into target work.

The skill is written to trigger on plain-language causal questions (“is LDL cholesterol actually causal for heart disease?”, “does BMI cause type 2 diabetes or just correlate?”) without the user naming MR at all.

Primary use cases: genetic validation of a drug target, testing whether a biomarker is a causal risk factor, triangulating an observational epidemiology result.

Notes

It is a reasoning layer over ToolUniverse; without the MCP server registered, the tool calls fail. Two scope exclusions are explicit upstream: plain GWAS association lookups belong to the GWAS skills, and the skill will not fit your own instruments from raw summary statistics you supply — it works from OpenGWAS study identifiers.

Genetic correlation is deliberately reported as not causation; treat that output as context rather than evidence. For association-level questions use GWAS Drug Discovery or the GWAS Catalog; for efficacy evidence assembly see Drug Target Validation. ToolUniverse ships ~68 such skills; other workflows are catalogued separately.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.