Today’s network rebuild is unusual. The provider count moved by 8, the API count by 52 — both modest. But the capabilities index expanded from 2,263 to 32,565 entries, a 14× deepening that reshapes how the catalog can be searched, filtered, and reasoned about.
This isn’t growth in coverage. It’s growth in granularity. The same providers, indexed at a much finer resolution. The rest of the catalog — Providers, APIs, Schemas, Rules — moved in normal increments. The story is in what happened to capabilities.
What Changed
For most of the catalog’s history, a single provider’s capabilities lived in one capabilities/<provider>-<capability>.yaml file at the upstream api-evangelist mirror. 1Factory had one capability — “Quality Management.” Most providers worked the same way: one capability per provider, summarizing what the provider’s APIs do at a high level.
The upstream capability files have now been split. Each capability YAML now describes a single coherent resource cluster or operation surface, not the whole provider. 1Factory now has three: 1factory-customers, 1factory-manufacturing, and 1factory-part-master — one for customer-side operations, one for inspection workflows, one for the part-master catalog. The same is true across thousands of providers.
The result, in the rebuild stats:
| Metric | Yesterday | Today | Delta |
|---|---|---|---|
| Providers | 4,411 | 4,419 | +8 |
| APIs | 16,219 | 16,271 | +52 |
| Capabilities | 2,263 | 32,565 | +30,302 |
| Category implementations | 298 | 1,913 | +1,615 |
| Schemas | 42,419 | 42,444 | +25 |
| Rules | 1,348 | 1,349 | +1 |
The category aggregator at capabilities.apis.io/categories/ reflects the deepening directly. The 26 canonical capability buckets — payments, monitoring, iot, object-storage, and the rest — now have 1,913 implementations mapped to them, up from 298. Mastercard alone now contributes capability nodes for card issuance, fraud, settlement, rewards, and tokenization separately, instead of one rollup entry.
Why This Matters
Coarse capabilities are easier to publish but harder to reason about. “1Factory does Quality Management” is true at a high level but doesn’t help an agent or developer answer questions like “which providers expose part-master lookup APIs?” or “which catalogs let me query manufacturing inspection records?”. Those questions live one level below the capability headline, and until today the catalog couldn’t answer them.
The new finer-grained index makes capability-level matching meaningful in three places:
-
Search by what you want to do. The search experience at apis.io now sees
manufacturing-inspectionsandpart-master-lookupas distinct capabilities, rather than collapsing them under one provider summary. The signal-to-noise for a query like “inspection record API” goes up substantially. -
Cross-provider capability matching. The 1,913 implementations now mapped to the 26 canonical capability buckets means the category pages function more like genuine cross-provider directories. The
paymentspage sees Stripe’s individual payment surfaces, Mastercard’s, Adyen’s, and so on — at the level where someone evaluating a payment integration actually compares them. -
Agent-readable specifics. Agents fetching capability profiles get specifics they can act on — operation lists, tool catalogs, resource paths — at the resolution that maps to actual integration work. The MCP surface gets meaningfully richer without the providers shipping more APIs.
The Cleanup Underneath
Today’s rebuild also fixed a backlog of upstream issues that had been quietly accumulating:
- 868
apis.ymlfiles in the upstream mirror had an indentation regression — first item incommon:correctly at 2 spaces, subsequent items dropped to column 0, breaking YAML parsing. Likely the residue of an automated property-adder run. Fixed in-place via a new fix-apis-yml-indent.py helper. - The category-suggestions seed file was generated against the old coarse capability schema and didn’t match the new finer slugs, which caused the category aggregator to collapse from 26 to 1 in an intermediate build. Regenerated suggest-categories.py against the new mirror — 35,116 candidate suggestions, 1,913 above the score threshold.
build.pyis now resilient to per-file YAML errors. A single malformed file in the upstream mirror logs a warning and continues; previously it killed the whole rebuild. Worth doing for any pipeline that pulls from many independent sources.- A handful of stray YAML issues in upstream vocabulary and ruleset files (coveo, 9 Spectral ruleset files) noted for separate cleanup.
New Providers
Eight new providers landed in this rebuild:
- ESRI ArcGIS — Geographic information system platform.
- EU Open Data Portal — Single point of access for open data across European Union institutions and member states.
- FIS — Financial technology infrastructure for banking and payments.
- Granular — Agriculture business management.
- Guidewire — Insurance platform.
- IFS — Enterprise resource planning.
- Infor — Industry-specific ERP and supply chain.
- LabVantage — Laboratory informatics.
The mix is itself interesting — three ERP/enterprise platforms (IFS, Infor, Guidewire), one mapping platform (ESRI), one EU public-data portal, one agriculture vertical (Granular), one fintech (FIS), one lab-informatics (LabVantage). All eight are domains where the new fine-grained capability schema actually pays off — each of these providers has many distinct operation surfaces that benefit from being indexed separately rather than under a single “Enterprise X Platform” headline.
What’s Next
The capability deepening is the structural change for this quarter. Three follow-ups are worth watching:
- Operations-level indexing. The next layer down — individual REST operations and MCP tools — already exists inside each capability file but isn’t yet surfaced as its own searchable index. The natural extension is to expose
apis.io/operations/{slug}URLs the same way capabilities, providers, and APIs are exposed now. - Capability-to-capability relationships. With 32,565 distinct capability nodes, the value of “which other capabilities are commonly combined with this one” goes up substantially. The vocabulary layer at
vocabularies.apis.ioalready tracks 3,207 resources across the catalog; cross-linking is the natural next step. - Re-scoring the suggestions. The new 35,116-row category-suggestions file has many low-confidence matches. A second pass tuned against the finer-grained capability schema would lift the implementation count above 1,913.
Every new provider and every new capability is now live across the network — at providers.apis.io, apis.apis.io, capabilities.apis.io, and the rest of the subdomain set. The federated api-catalog feed picked up all 52 new APIs in the same rebuild.
If you ship an API and we should be indexing it, the upstream lives at api-evangelist — open a PR with an apis.yml and the next rebuild picks it up.