Say you run a small publication on Ghost, and you want an agent to draft and publish posts for you. Before you let it near your site, you open the APIs.io record for Ghost to see what it would be working with. The first thing you see is a long list of API pages. Tags API sits next to Ghost Admin - Tags API, and nothing tells you which one is current.
The post you are reading says that confusion is the catalog's fault, not Ghost's.
Start with the score. Ghost sits at 77.1 on the Kin Score, in the exemplar band. That is down 1.3 from 78.4. Its Agent Readiness is 30.8, agent-ready. The post calls this a strong record for an MIT-licensed publishing platform that funds a non-profit foundation, and says most of it is earned by Ghost itself.
Here is what you would actually be calling. Every Ghost site exposes two documented REST APIs under https://{site}/ghost/api/. The Content API is read-only and key-authenticated, and it delivers posts, pages, tags, authors, tiers and settings to front-ends. The Admin API is read-write and JWT-authenticated, and it manages posts, pages, members, tags, tiers, offers, newsletters, users, media, themes and webhooks. If your agent is going to publish for you, the Admin API is the one that matters.
Now the part that is not clean. The provider record says three source specs were split into 33 pages. By tag, those three files became six Content pages holding 14 operations, thirteen admin-prefixed pages holding 42, and fourteen unprefixed pages that cover the same Admin resources again with 48. The agentic-access artifact counts 56 operations, 26 of them acting, which is exactly 42 plus 14. So the real surface is 56 operations. The fourteen unprefixed pages are a second copy of the Admin API from an earlier split that was never retired. The post says plainly that this defect is the catalog's, not Ghost's.
The facet scores tell you where Ghost is strong. Operational transparency is 89.5 and developer ergonomics is 86.9. Discoverability is 80.0 and access clarity is 76.3. Contract quality is 66.7. Contract governance is 31.8, the weak column, and the upsert check scores 0. The open-source facet scores 100. All 33 contracts score as callable even though they are templated on the customer's own domain. The post calls that honest, because every Ghost install is its own host.
On the agent side, Ghost ships first-party MCP. A hosted, anonymous, read-only documentation server runs at docs.ghost.org/mcp and answers a live tools/list. A separate local server built into ghst, Ghost's CLI, exposes the Admin API as roughly 90 tools once an operator has authenticated it. That local server is the one you would care about for publishing.
Then comes a second admission from the catalog. It probed an A2A agent card at docs.ghost.org/.well-known/agent-card.json on 2026-09-26 and graded it conformant at protocol 0.3. Yet the Agent Readiness record says agent_card: false. Either the dimension is measuring a different host than the probe found, or the probe result never reached the scorer. Either way, Ghost published the artifact and the score does not show it. The post files this as a catalog defect, not a Ghost gap.
Some dark dimensions are real, and you should weigh them. There is no idempotency, no protected resource metadata, no dynamic client registration, no reversibility notes, and no examples in the contracts. The post does not credit Ghost for agentic access, because that is derived.
So what do you take away? The surface you would hand to an agent is 56 operations, not 33 pages' worth. The duplicate Admin pages are noise the catalog owes you a fix for, and it says it will retire the fourteen and reconcile the agent card. On Ghost's side, the post names three things that would move the number: publishing its own OpenAPI for both APIs, adding request and response examples, and documenting idempotency on the Admin writes. That last one matters if an agent is publishing posts or changing members, because a retried write is the risk you would be taking on. The documentation and the open-source work are already exemplar. The contract layer is what is left.
Read the original: https://apis.io/2026/10/26/ghost-scores-77-on-fifty-six-operations-the-catalog-lists-twice/