Turn the GWAS Catalog into a searchable evidence base.
Find relevant studies, filter by sample size and ancestry, inspect top associations, and compare results — directly from your AI agent.
Built from the EMBL-EBI GWAS Catalog, enriched by Devano, and linked back to the source at every study.
Snapshot from corpus_overview · August 12, 2026.
GWAS evidence is abundant. Using it should be easy.
Same underlying evidence — restructured into a form you can actually query.
From a question to usable evidence.
Four things the index makes routine. Every result below is a live call against the index, shown as returned.
Search by meaning, not exact catalog wording — two loose words find the right disease, cleanly split from the 72% of the catalog that's molecular and quantitative-trait scans.
Ranked by relevance, not string position. The chips are MONDO terms —
MONDO_0005016,
MONDO_0005300 — so one query
reaches every study mapping that disease, regardless of phrasing.
Filter and sort by parsed sample size, case/control counts, ancestry, and trait category — sample descriptions that used to be prose, now numbers you can actually use.
"39,106 European ancestry clinically diagnosed cases, 46,828 European ancestry proxy cases, 401,577 European ancestry controls"
Not filterable. Not sortable. Not comparable.
Filter. Sort. Aggregate.
92 well-powered Alzheimer's studies, ranked and comparable in one call — GW-significant counts computed by Devano, not recorded in the catalog.
Lead variants, p-values, effect sizes, and nearest genes — grounded to dbSNP and HGNC. 91.6% of studies carry a top-hit summary, computed from deposited summary stats even when the record has none.
| variant | nearest gene | context | p-value | OR |
|---|---|---|---|---|
| rs6733839 | BIN1 | intron | 6×10−118 | 1.17 |
| rs3851179 | intergenic | — | 3×10−48 | 0.90 |
| rs679515 | CR1 | intron | 7×10−46 | 1.13 |
| rs11787077 | CLU | intron | 2×10−44 | 0.91 |
| rs1582763 | MS4A4A | intergenic | 4×10−42 | 0.91 |
| rs12151021 | ABCA7 | intron | 2×10−37 | 1.10 |
| rs75932628 | TREM2 | missense | 3×10−37 | 2.39 |
Every variant links to dbSNP, every gene to HGNC — grounded to canonical identifiers, so a gene filter matches the gene, not every label that contains its letters.
Live call · study_detail(study_accession="GCST90027158")
Compare two studies, or a paper's sibling accessions — sixteen independent replications, or one paper sliced sixteen ways. Read as replications, those slices make a target look far better supported than a single paper backs. See who actually did the work: first and last author, with profile links, on 99.6% of studies.
Early vs. late disease, ESRD vs. no ESRD, all-diabetes vs. type 2, plus glomerular filtration rate — sixteen accessions that look independent, but aren't.
Resolved from the publication itself, not matched from a surname — each with an OpenAlex profile and affiliation.
Designed for scientific agents.
Ask in plain language, get an answer — not a data dump to sift. An agent orients, searches, opens only what matters, then compares, so its context fills with findings instead of raw records.
One MCP. Four resolutions. Orient → search → inspect → compare.
Traits, genes, variants, and authors resolve to shared public identifiers — results join cleanly with public resources and your own systems.
Precomputed, no model in the query path — predictable latency and reproducible output, built to sit inside an automated workflow.
Enriched, but never disconnected from the source.
GWASx sits on the EMBL-EBI GWAS Catalog — every study links back to its
canonical entry, and every response labels what's derived or computed via a
devano_derived list. The
catalog is the authoritative, live source; GWASx is the enriched layer on top,
rebuilt on deterministic pipelines to track it.
The GWAS Catalog record, unaltered — accession, publication, platform, curated associations, and the catalog's own trait mapping.
Trait category and group, parsed sample sizes, primary ancestry, design hint, grounded genes and authors — each listed by name in the response.
Top-hit summaries calculated from deposited summary statistics: genome-wide-significant counts, lead variants, and nearest genes.
GWASx reports lead associations — the strongest signal in a region, not the fine-mapped causal variant — so nearest-gene annotations mark location, not mechanism. No deposited summary statistics? We say so, not blank. Every study links back to its GWAS Catalog record, tracked as the catalog updates.
Start exploring human genetics evidence.
claude mcp add --transport http \ devano-gwas-index https://devano.ai/mcp/gwas-index
codex mcp add devano-gwas-index \ --url https://devano.ai/mcp/gwas-index codex mcp login devano-gwas-index
Works with Claude, Codex, and any MCP-compatible client — read-only and precomputed. Your agent completes a one-time sign-in on first use.
All 186,237 studies. No trimmed corpus, no capability held back.
Categorization, parsed sample sizes, grounded traits, genes and authors, and computed top hits.
Persistent history, larger quotas, and saved searches.
Headline counts: corpus_overview
on the GWAS Index build, snapshot 2026-08-12. Per-study examples:
search_studies,
study_detail, and
paper_overview.
Top-hit summaries come from GWAS Catalog-deposited summary statistics —
variants filtered at genome-wide significance (p < 5×10−8),
clumped by distance to the lead variant, and annotated with the nearest gene. Trait
grounding: EFO/MONDO. Gene grounding: HGNC. Author grounding: OpenAlex. Numbers
refresh with each rebuild as the catalog updates.
Source data is the EMBL-EBI GWAS Catalog; underlying studies are credited to their original authors and journals. Enrichment presented on this page is provided by 🧬 Devano (devano.ai).