Grafana publishes 138 APIs on the network with a composite of 72.7 — exemplar — and an agent-readiness score of 52.3 (agent-ready). Its agentic-access contract enumerates 378 operations, 204 acting, 24 human-in-the-loop. It is freemium with self-serve signup, and you can try it without talking to anyone.
Two numbers in that paragraph do not look like the others: 138 APIs and 378 operations.
Under three operations per API
GitHub averages 6.3 operations per API entry. Telnyx, 6.2. Grafana averages 2.7.
That is what happens when a surface is split by resource all the way down — Access, Account, Accounts, Add, Address, Administrative, Alerts, Annotations, Cache, Calls, Checks, Cloud, Comments, Connections, Convert, and 123 more. Some of these are two-endpoint definitions.
I am of two minds about it. Fine-grained splitting is genuinely better for retrieval: an agent looking for annotation management finds a document about annotations rather than a 4,000-line monolith it has to search. But a 138-entry surface where the entries are named Add, All, and Clean has crossed from resource decomposition into artifact sprawl, and the naming is doing no work for a human or a model trying to pick one.
The facet spread
| Facet | Score |
|---|---|
| Discoverability | 85.2 |
| Commercial clarity | 78.9 |
| Contract quality | 72.3 |
| Governance | 69.8 |
| Operational transparency | 68.4 |
| Developer ergonomics | 65.2 |
This is the flattest facet profile of any provider in this week’s sweep — 20 points between best and worst, no zeros, nothing below 65. Compare Telnyx, profiled yesterday: 64 points of spread and a facet at zero.
Governance at 69.8 is what open source buys you. When the API definitions live in the same repository as the code, under the same review process, “how is this governed” has a public answer. Most closed-source vendors of comparable size score half this.
Discoverability at 85.2 is the other open-source dividend. The specs are on GitHub, findable, forkable, and not behind a developer portal login.
Every contract is callable
The provenance block records 138 contracts, 0 derived, 100% callable. Every specification resolves to a real production host — no example.com placeholders, no bare {apiRoot} templates.
That sounds like a low bar and is not. Across the catalog, 10,950 of 87,612 specs are uncallable as published. Grafana’s 100% is a direct consequence of the definitions being the ones the product actually uses, rather than a documentation artifact maintained alongside the real thing and drifting from it.
The MCP server is recorded as first-party — Grafana’s own, not one derived on its behalf. That distinction matters to the score and it should: an agent interface a vendor maintains is a commitment, and one generated for them is a convenience.
Takeaway
138 APIs, 378 operations, 100% callable contracts, and the flattest facet profile in the sweep. Grafana’s score is the open-source dividend made legible — governance and discoverability come nearly free when the spec lives with the code. The one thing worth fixing is nomenclature: nobody can pick between Add API and All API.
See the full profile at apis.io/providers/grafana/.