This Kubernetes MCP server has 22 READ_ONLY tools covering cluster introspection. Major issues: (1) 9 of 22 tools lack descriptions entirely (list_nodes, list_services, list_deployments, list_ingresses, list_persistent_volumes, list_persistent_volume_claims, list_pods, list_events, list_config_maps, list_secrets, get_pod, get_service, get_node). (2) Input schemas are present only for tools with explicit parameters documented; many list_* tools lack input schema visibility. (3) Parameter descriptions are minimal or absent for list_* operations. (4) No output schemas documented anywhere. (5) No error handling guidance, recovery hints, or actionable error messages shown in source. (6) Tool names follow verb_noun convention (good), but lack context about what distinguishes list_pods from get_pod or whether pagination is supported. Average tool description length is ~40 chars for those with descriptions, below the 194-char baseline for A+ tools. Conservative scoring reflects these gaps.

• Add descriptions to all 11 list_* tools. Each should explain: 'Lists all [resource type] in the cluster (or namespace if namespace-scoped). Returns a summary of each resource. Call this first to discover available resources before calling get_[resource].'
• Add descriptions to get_service, get_pod, get_secret following the pattern: 'Retrieves detailed information about a specific [resource] by name and namespace. Use after list_[resource] to fetch full details for a specific item.'
• Document input schemas for all list_* tools. Include parameters like namespace (string, optional), label_selector (string, optional), limit (integer, 1-500, default 20), continue_token (string, optional) if pagination is supported.
• Document output schemas for all tools. Example for list_pods: 'Returns {items: [{name: string, namespace: string, status: {phase: string, ...}}, ...], total: integer, continue_token?: string}'.
• Add error handling guidance to get_* tools: 'If the resource is not found, returns an error. Call list_[resource] to discover available resources and check spelling/namespace.'
• Enforce result limits in list_* tools (cap at 20-50 items by default) to avoid context explosion. Document in descriptions: 'Results are limited to 50 items; use limit and continue_token for pagination.'
• Add parameter descriptions for namespace parameters in get_deployment, get_ingress, get_persistent_volume_claim, get_config_map to clarify 'The Kubernetes namespace where the resource resides (default: default).'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry