Hygraph

Content API: POST to the project's GraphQL endpoint with queries and mutations. Auto-generated queries for each model: model (single by ID/unique field), models (list with filtering), modelsConnection (paginated with aggregate counts). Auto-generated mutations: createModel, updateModel, deleteModel, publishModel, unpublishModel. Filtering uses where input types with field-specific operators. Ordering, pagination (first/skip or cursor-based), and localization (locale parameter) are built into the query schema. Asset management uses the Asset model with upload mutations. Webhooks deliver events for content lifecycle changes. The Management API provides schema operations (model and field CRUD) via GraphQL. The pure GraphQL approach is elegant for agents using GraphQL tooling but adds friction for agents preferring REST.

Hygraph (formerly GraphCMS) is built entirely around GraphQL — there's no REST API. For agents, this means all content operations use GraphQL queries and mutations. The schema is defined visually or via the Management API, and the Content API auto-generates typed GraphQL operations. Content federation is unique to Hygraph: it can combine CMS-managed content with data from external APIs (REST, GraphQL, or databases) in a single GraphQL schema. For agents building content-heavy applications that need to merge CMS data with external data, federation eliminates the need for a separate orchestration layer. Localization is built into the schema with per-field locale support. Stage-based publishing (draft, published, custom stages) enables editorial workflows. For agents comfortable with GraphQL, Hygraph provides a powerful content API.

Permanent Auth Tokens (PATs) provide API access with configurable permissions: content read, content write, management read, management write. Tokens are per-environment. Public Content API access can be enabled for unauthenticated reads (CDN-cached). For agents, creating a content write PAT for content management and using public access for reads provides clean separation. No OAuth for third-party integrations. Tokens don't expire. The permission model is more granular than Sanity's (content vs. management separation) but less granular than Contentful's (no per-model scoping). For agents, the PAT model is straightforward. IP-based restrictions are available on enterprise plans.

Documentation at hygraph.com/docs covers API reference, schema management, content operations, and integration patterns. The GraphQL API playground (built into the dashboard) enables interactive query construction and testing. Schema documentation is auto-generated from the content model. SDK: @hygraph/management-sdk for schema operations, generic GraphQL clients (graphql-request, Apollo, urql) for content operations. The documentation assumes GraphQL familiarity — it doesn't teach GraphQL fundamentals. Content federation documentation explains how to connect external data sources. For agents, the API playground and the auto-generated schema documentation are the most valuable resources — they enable schema-aware query construction.

GraphQL errors return in the standard errors array with message, locations, and extensions. Validation errors include field-specific details. Rate limits vary by plan: community plan has tighter limits, paid plans are more generous. The GraphQL query complexity is measured and limited — overly complex queries (deep nesting, wide selections) are rejected to prevent abuse. Content publishing validation ensures required fields are populated before publishing. Webhook delivery retries on failure. The main consideration for agents: GraphQL query optimization matters more than with REST — poorly structured queries can hit complexity limits or perform slowly. Agents should request only needed fields and use pagination appropriately.