GraphQL’s promise to an integrator, and now to an agent, is that the schema is the contract. You introspect it, you see every type and field, and you ask for exactly what you need. So the useful question about the GraphQL index on apis.io is not how many providers offer GraphQL. It is how many of those schemas you can actually read from the catalog. The answer, from the index itself, is none.
What the index holds
The GraphQL collection has 1,016 records. 661 of them are redirects to a provider page. That leaves 355 live GraphQL entries across 352 providers. Each one is a descriptor: a name, a description, an endpoint, a docs URL and reference links.
| Field | Entries carrying it |
|---|---|
| Documentation URL | 262 |
| Endpoint | 222 |
| Both | 213 |
| Neither | 84 |
| A schema (SDL or introspection result) | 0 |
84 entries tell you a provider has GraphQL and give you no endpoint and no docs URL to find it. Of the reference links across all entries, only 6 are typed as a schema. The spec_file on every record is the Markdown descriptor itself.
The schemas exist, one layer down
The provider repositories under the index tell a different story. 394 .graphql files sit in the graphql/ folders of 308 provider repos, with a median of 1,266 lines. Their provenance is mixed, and the headers of the files I sampled say so: some are the provider’s own schema copied from its public repository, some are introspection output from a provider’s starter template, and some are modeled from the provider’s API reference by us. The index surfaces neither the file nor that distinction.
That is the gap. An integrator lands on a GraphQL entry and gets a pointer. The artifact that makes GraphQL worth choosing sits in a repo, unlabeled as to whether the provider wrote it.
Who publishes GraphQL, and how they score
GraphQL providers score well above the catalog. The 351 scored providers with a GraphQL entry have a median Kin Score of 36.5, against 14.3 across the 27,360 scored providers catalog-wide. 32 are exemplar and 35 strong. The top of the list is dominated by platforms that run GraphQL beside REST: Salesforce at 91.8, Tray.ai at 89.1, Shopify at 83.9, Appwrite at 83.2 and Optimizely at 80.8. In fact, 223 of the 355 entries also list an OpenAPI. GraphQL on this network is mostly a second interface on a mature platform, not a replacement for the first.
The Kin Score reads a provider’s public surface, and a derived schema is not credited to anyone. That is correct, and it is also why the labeling matters: a provider that publishes its SDL deserves to be told apart from one whose schema we reconstructed.
What a good entry looks like
Endpoint, docs URL, and a link to the SDL or introspection result the provider itself publishes, marked first-party or derived. Add the authentication scheme and whether introspection is enabled in production, since many providers turn it off and an agent needs to know before it tries. That is five fields, and today the best entries have three.
Takeaway
The GraphQL index is a good directory and a poor contract library. The fix is on both sides: providers should publish their SDL at a stable URL, and the index should show it with its provenance. Until then, browse the 352 providers at apis.io/graphqls/ and treat each entry as a starting point, not the contract.