release: 2.3.0 — add the NHGRI-EBI GWAS Catalog (37th database)

CHANGELOG.md

@@ -15,6 +15,41 @@ dominant client re-reads the schema each session. Only a removal/rename is MAJOR
 
 _Nothing yet._
 
+## [2.3.0] - 2026-07-31
+
+<!-- whatsnew: 2026-07-31 | New database: the <strong>NHGRI-EBI GWAS Catalog</strong> — 955,930 SNP–trait associations across 89,981 studies, with p-values, effect sizes, risk alleles and mapped genes, joinable to EFO trait terms. Ask things like "which variants are associated with QT interval, and how strong is the evidence?" -->
+
+### Added
+
+- **`gwascatalog` — the NHGRI-EBI GWAS Catalog, the 37th database.** Published genome-wide
+  association results: 955,930 SNP–trait associations (801,096 at genome-wide significance) over
+  444,106 SNPs, 89,981 studies and 11,017 EFO traits. It sits on the `ebi` endpoint we already talk
+  to, so this is a registry row plus an MIE — no new infrastructure.
+  Picked from the production tool-call log rather than by guesswork: with the ~94% of traffic that is
+  a load-test harness and one automated pipeline filtered out, the largest remaining usage cluster is
+  disease-variant work chaining `clinvar` → `togovar` → `medgen` → `mondo` → `hco` → `hgnc`, and
+  trait association is the layer that thread was missing.
+  The MIE ships 10 live-verified examples (4 basic, 3 intermediate, 2 aggregation, 1 cross_db) and
+  documents three traps an agent cannot recover on its own. **The two vocabularies in the graph store
+  string literals in opposite term forms** — `gwas:has_snp_reference_id` needs `^^xsd:string` while
+  `m2r:snps` needs the plain form, and `DATATYPE()` reports `xsd:string` for both, so only an `ASK`
+  tells them apart; getting it backwards returns 0 rows silently. **Two parallel association models**
+  coexist: 955,930 med2rdf blank-node `Association`s carrying the flat row, versus 782,879
+  IRI-addressable OBAN `TraitAssociation`s — different populations, and mixing them double-counts.
+  And **`?s a gwas:Study` returns 224,583 across two disjoint IRI families**, of which only 89,981 are
+  real GCST study records; the rest are EBI `Trackable/*` publication stubs.
+  Trait labels live only in the co-hosted `ontology/efo` graph, so the `cross_db` example is the
+  route to trait names, not a nicety.
+
+### Changed
+
+- Adding a database is now documented as touching **five** files, not one: `endpoints.csv`, the MIE,
+  the regenerated `02b_database_catalog.md`, the **hand-written** endpoint table in
+  `02_budgets_and_discovery.md` (no generator touches it — two `TestUsageGuideEndpointTable` tests
+  are what catch a miss), and the intro page's database grid. The README's **Contributing** section
+  now lists all five with the guarding test for each; it previously said two, which is how three of
+  them got missed on the first pass of this very release. A *removal* really is just the registry row.
+
 ## [2.2.1] - 2026-07-30
 
 ### Added
@@ -751,7 +786,8 @@ their own file. No tool-surface change; the served MIE/guide content is correcte
 _MIE database onboarding and revisions land continuously and are summarised per
 release above; see git history for the full detail._
 
-[Unreleased]: https://github.com/dbcls/togomcp/compare/v2.2.1...HEAD
+[Unreleased]: https://github.com/dbcls/togomcp/compare/v2.3.0...HEAD
+[2.3.0]: https://github.com/dbcls/togomcp/compare/v2.2.1...v2.3.0
 [2.2.1]: https://github.com/dbcls/togomcp/compare/v2.2.0...v2.2.1
 [2.2.0]: https://github.com/dbcls/togomcp/compare/v2.1.3...v2.2.0
 [2.1.3]: https://github.com/dbcls/togomcp/compare/v2.1.2...v2.1.3

README.md

@@ -175,7 +175,7 @@ TogoMCP exposes tools for querying the following (via SPARQL or REST APIs):
 | Category | Resources |
 |---|---|
 | Proteins / Proteomics | UniProt, PDB, jPOST |
-| Genes / Genomics | NCBI Gene, Ensembl, HGNC, OMA, Bgee, HCO, MCO, DDBJ, MoG+, TogoVar |
+| Genes / Genomics | NCBI Gene, Ensembl, HGNC, OMA, Bgee, HCO, MCO, DDBJ, MoG+, TogoVar, GWAS Catalog |
 | Chemistry | ChEMBL, PubChem, ChEBI, Rhea, BRENDA, MassBank |
 | Pathways | Reactome |
 | Disease / Clinical | ClinVar, MedGen, MONDO, NANDO |
@@ -232,7 +232,15 @@ togomcp/
 
 Contributions are welcome!
 
-**Adding a database**: add an MIE file under `togo_mcp/data/mie/` and a corresponding row in `togo_mcp/data/resources/endpoints.csv` (see the MIE spec in `togo_mcp/data/docs/`).
+**Adding a database**: five places, not two. Only the first two affect what the server *validates*; the rest are documentation surfaces that drift silently, and the tests are what catch them.
+
+1. `togo_mcp/data/resources/endpoints.csv` — the registry row (this alone decides valid `database=` values).
+2. `togo_mcp/data/mie/<db>.yaml` — the MIE file (see the MIE spec in `togo_mcp/data/docs/`).
+3. `uv run python scripts/generate_usage_guide_catalog.py` — regenerates the Usage Guide's database catalog. Guarded by `tests/test_catalog_in_sync.py`.
+4. `togo_mcp/data/resources/usage_guide_v6/02_budgets_and_discovery.md` — a **hand-written** copy of the registry that no generator touches. Bump the per-endpoint count *and* add the key. Guarded by `TestUsageGuideEndpointTable` in `tests/test_server.py`.
+5. `togo_mcp/data/docs/togomcp-intro.html` — add a card to the database grid (not generated).
+
+Note that a database *removal* really is just step 1: nothing validates against the other four.
 
 **Adding a tool**: pass `annotations=READ_ONLY_TOOL` to the `@mcp.tool` decorator. Every TogoMCP tool is read-only, and MCP's default for an *unannotated* tool is the unsafe one — clients such as ChatGPT treat a tool with no `readOnlyHint` as a write action, which means a confirmation prompt on every call. A test asserts this, so omitting it fails the build.
 

pyproject.toml

@@ -15,7 +15,7 @@ togo_mcp = [
 
 [project]
 name = "togo-mcp"
-version = "2.2.1"
+version = "2.3.0"
 description = "MCP Servers for using the RDF Portal"
 authors = [
     {name = "Akira R. Kinjo"},

togo_mcp/data/docs/togomcp-intro.html

@@ -767,6 +767,10 @@ <h2 class="section-title">What's New</h2>
     </div>
     <ul class="whatsnew-list">
       <!-- WHATSNEW:START -->
+      <li>
+        <span class="whatsnew-date">2026-07-31</span>
+        <span>New database: the <strong>NHGRI-EBI GWAS Catalog</strong> — 955,930 SNP–trait associations across 89,981 studies, with p-values, effect sizes, risk alleles and mapped genes, joinable to EFO trait terms. Ask things like "which variants are associated with QT interval, and how strong is the evidence?"</span>
+      </li>
       <li>
         <span class="whatsnew-date">2026-07-30</span>
         <span>Drug lookups in ChEMBL now find <strong>biologics</strong> — antibodies, therapeutic proteins, vaccines and cell therapies were silently missing, so names like <em>Rituxan</em> or <em>efalizumab</em> returned nothing. A new <code>mode='extract'</code> also resolves the drugs named inside a clinical-trial intervention string such as "Ropivacaine 10% + Clonidine".</span>
@@ -783,10 +787,6 @@ <h2 class="section-title">What's New</h2>
         <span class="whatsnew-date">2026-07-17</span>
         <span>Query results are more trustworthy: <code>get_MIE_file</code> now leads with each database's "traps" (mandatory filters, namespace pitfalls, and shared-graph count inflation), and all 36 databases document cross-graph co-tenancy so counts aren't silently inflated.</span>
       </li>
-      <li>
-        <span class="whatsnew-date">2026-07</span>
-        <span>Published in <em>Database</em> — <a href="https://doi.org/10.1093/database/baag042" target="_blank" rel="noopener">2026:baag042</a>.</span>
-      </li>
       <!-- WHATSNEW:END -->
     </ul>
     <p class="whatsnew-more"><a href="https://github.com/dbcls/togomcp/blob/main/CHANGELOG.md" target="_blank" rel="noopener">Full changelog →</a></p>
@@ -1121,6 +1121,10 @@ <h2 class="section-title">Available Databases</h2>
         <div class="db-card-name">MoG+ (Mouse Genomes plus)</div>
         <div class="db-card-desc">Genome-wide sequence variation across 62 inbred and wild-derived mouse strains on the GRCm39 assembly: ~146M variants (SNVs and indels) with per-strain genotypes, Ensembl VEP and SnpEff functional consequences, and Ensembl gene-overlap links.</div>
       </div>
+      <div class="db-card">
+        <div class="db-card-name">GWAS Catalog</div>
+        <div class="db-card-desc">NHGRI-EBI catalog of published genome-wide association studies: 955,930 SNP–trait associations (801,096 at genome-wide significance) across 89,981 studies and 444,106 SNPs, with p-values, effect sizes, risk alleles, mapped genes, and 11,017 EFO trait terms joinable to the co-hosted EFO ontology.</div>
+      </div>
       <div class="db-card">
         <div class="db-card-name">TogoVar</div>
         <div class="db-card-desc">Integrated human genome variation database (GRCh38): ~390M variants (SNV, deletion, insertion, MNV, indel) with normalized/VCF coordinates, Ensembl-VEP consequences (SIFT/PolyPhen/AlphaMissense, HGVS), dbSNP and HGNC/Ensembl gene links, and an embedded ClinVar annotation layer. Per-population allele frequencies are served via the TogoVar REST API tools.</div>