Hygraph's agent door scores twice its developer door

Hygraph's agent door scores twice its developer door

Hygraph has put its MCP connector in the Claude Directory, and the announcement makes a bigger claim than a connector usually carries. In Announcing Hygraph’s MCP Connector for Claude, Paul Biggs calls it “another step in the rise of the Double Headless CMS,” the idea that headless decoupled the experience layer a decade ago and “left everything below the surface intact… including one interface nobody thought to question: our own.” Connecting takes a click: in Claude, open Connectors, find Hygraph, sign in to authorize a project. From there an agent can ingest content from Google Docs or Notion, bulk-update entries, create entries across locales with relationships, inspect and refactor the schema, and generate GraphQL queries. The sentence that earns the post is about the last of those: “An agent that can shape a knowledge graph is doing content architecture, not autocomplete.”

There are no numbers, only customer voices, and they are the marketing. “MCP is my new best friend” from a services partner, and a digital lead who says “I basically want everybody in Marketing to be able to almost… not open Hygraph anymore.” That second quote is the one that matters, because it is the whole thesis said by a buyer: the product’s own interface is now optional for a class of users. What survives independent of the enthusiasm is the shape of the connector. It is scoped and permissioned per project, authorized through the standard Claude connector flow, and it exposes both content operations and schema operations, which is the part that turns a writing assistant into something that can restructure the model everything else depends on.

The catalog agrees the door exists and has the measurements to say something about it. The Hygraph provider page lists 2 API pages: the Hygraph GraphQL Content API, which the connector’s content operations run against, and the Hygraph Management API, which its schema, component, and enum operations run against. On the Agent Readiness score the MCP server is lit, and so is the whole identity layer: delegated identity, protected resource metadata, and dynamic client registration, which is exactly what a directory-listed connector with a sign-in flow has to publish.

The Kin Score is 22.2, emerging band, and that is the finding. Discoverability carries it at 60.0 and contract quality at 39.4, but access clarity is 10.5, developer ergonomics is 11.9, operational transparency is 18.4, and contract governance is 0.0. The Agent Readiness score is 42.8, agent-ready, nearly twice the composite. Hygraph has built an agent entrance that is better described than its developer entrance: a human arriving at the record cannot easily find how to get access, what the errors look like, or how the contract changes, while an agent arriving through Claude finds a registered, permissioned, OAuth-described door. The Double Headless thesis says the CMS’s own interface is optional now. On the catalog’s reading, Hygraph has already made the human one the harder of the two to walk through.

← Google judges an agent's intent at runtime, and the catalog cannot see the gateway
LaunchDarkly's Flag Triggers API mints a URL that flips production, and our page shows only two of its five operations →