Canonical scores 74.8 on the Kin Score, exemplar band, and it is falling: down 7.0 from 81.8 at the previous run. Its Agent Readiness is 43.3, agent-ready, and the rubric records that a gate held it back from agent-native. The company behind Ubuntu has one of the widest public API estates on the network. The record shows how little of that width an outside caller can actually reach.
What is in the surface
Canonical’s provider record says 13 APIs on a published basis: the Snap Store, Charmhub, MAAS, Juju, Launchpad, Ubuntu Pro, LXD, Landscape, Pebble, MicroCeph, the Ubuntu Security API, Testflinger and the Anbox Cloud surface. The catalog itself holds 135 API pages under the provider, and the score counts 117 contracts. Most of those pages are single-resource slices with names like Canonical Network Zones API, Canonical Placement Groups API and Canonical Sentry Debug API.
Of those 117 contracts, 12.0% are callable from outside. Much of the estate is built that way on purpose. LXD, MAAS, Pebble, Landscape and MicroCeph run on the customer’s own machine, socket or cluster, and the MAAS page says so honestly with a base URL of https://<maas-host>/MAAS/api/2.0. The catalog is less honest. 116 of the 135 API pages show https://api.snapcraft.io as their base URL, and that includes the network, cluster and instance slices that belong to LXD. That defect is ours, not Canonical’s. An agent that trusts those pages would send LXD calls to the Snap Store.
| Facet | Score |
|---|---|
| Access clarity | 92.1 |
| Developer ergonomics | 73.2 |
| Discoverability | 69.6 |
| Contract quality | 66.6 |
| Operational transparency | 60.5 |
| Contract governance | 35.6 |
Access clarity is the strongest facet. Most of these APIs are open, documented and sit next to their source code on GitHub (canonical/lxd, canonical/pebble, canonical/testflinger, canonical/identity-platform-api). Contract governance is the weakest. Marker coverage is 2.6%, which means almost none of the 117 contracts carry the versioning, lifecycle and ownership markers the rubric looks for. Under the Horizontal regulatory regime Canonical scores 45.1.
Where the agent record comes from
The Agent Readiness record has real first-party strengths. Error semantics are verified. Idempotency, reversibility, rate-limit signalling and delegated identity are documented. Launchpad publishes a webhook surface. Against that, there is no MCP server, no agent card, no well-known catalog and no protected resource metadata. The agentic-access artifact (812 operations, 413 acting, 3 human-in-the-loop) and the Agent Skills dimension are both marked derived: we built them, so they are not credited to Canonical here.
The seven-point drop needs a careful reading. The record shows the fall, but it does not show its cause. Before anyone calls it a regression at Canonical, it should be checked against the rubric version that scored the previous run.
Takeaway
Three things would move the number. Canonical could publish one canonical OpenAPI per product with lifecycle markers, so the catalog stops carrying 117 fragments under one wrong host. It could describe a reachable sandbox or public host for at least the Snap Store and Charmhub contracts, which already run as public services. And it could publish a well-known API catalog that ties the 13 products together. The documentation is already exemplar. The contracts are behind.
See the full profile at apis.io/providers/canonical/.