A container orchestration and management platform with MCP tools for querying and managing services, allocations, logs, events, statistics, pipelines, storage, backups, and audit logs. Supports planning and applying HCL job specifications.
Kanea demonstrates solid tool design with 14 well-named, read-only tools covering infrastructure inspection. All tools have descriptions (avg 150 chars, within 10-1024 baseline). Tool names follow verb_noun pattern (list_*, get_*). However, input schemas are not visible in the provided source code, only parameter names and descriptions are shown. This is a critical gap: without seeing actual JSON Schema definitions with type declarations, schema quality cannot be verified. The plan_spec tool lacks an input schema entirely. Error handling and output schemas are not documented. Descriptions are adequate but could be more prescriptive about when to call each tool vs. alternatives.
Read the audit log: who did what, when, from where, and whether it was allowed. Requires the admin role; the daemon enforces that.
Read the platform event feed newest-first: deploys, crashes, health transitions, scaling decisions, certificate issuance, builds.
Read the tail of a service's container logs. Not a stream: it returns the last N lines and comes back. Use it to diagnose a crash or a failing health check.
Summarise the node: version, how much is declared, how much is running, how much is failing, the machine's own CPU, memory and load, the metrics pipeline's health and whether the circuit breaker is open. A missing value means no reading, which is not the same as zero.
Describe one project: counts, git source, notification channels.
Input schemas not visible in source code. Cannot verify JSON Schema type definitions, constraints, or parameter validation.
Output schemas not documented. No evidence of return type definitions, field descriptions, or pagination guidance. LLMs cannot plan downstream tool calls without knowing what fields to expect.
plan_spec tool has no input schema shown and minimal description (70 chars). Unclear what HCL spec format is expected, what validation occurs, or what the output structure is.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Show one service's full declared state: image, count, environment variable names, resources, ports, health check, scaling policy and ingress. Environment values are shown; secrets are references, never values.
Current CPU, memory, request rate and p95 latency for a service and its allocs. A missing value means no recent data, which is not the same as zero.
List allocation (container) records with state, restarts and health. This is what is actually running, as opposed to what is declared.
List state archives newest-first, with when each was taken, the index it covers and what it holds. Also reports whether replication is actually working, which is the number that matters before an incident.
List build and deploy runs newest-first, with state, commit and duration.
List projects with their service and alloc counts, git source and configured notification channels. Start here when you do not know what is deployed.
List declared services with their image, replica count and resource limits. Optionally filtered to one project.
List the volumes services declare, with their backing storage type and which services mount them.
Parse an HCL job spec and report what applying it would change. Changes nothing. Always prefer this before apply_spec.
No error handling guidance documented. Tools do not indicate what errors are retryable, what user-fixable errors look like, or how to recover from failures (e.g., 'project not found').
Parameter descriptions lack format/constraint details. E.g., 'tail' param in get_logs says 'max 500' but no minimum stated; 'limit' params in list_* tools lack bounds. LLMs cannot validate inputs without explicit constraints.