The server defines 12 tools with generally good naming conventions (verb_noun pattern: select_, list_, flux_, kustomize_, helm_, kubeconform_, yaml_, argocd_). All tools have clear, substantive descriptions (194-280 chars average, well above baseline 34 minimum). However, schema completeness and parameter descriptions vary significantly. All tools shown have explicit input schemas with proper JSON Schema format and type definitions. Parameter descriptions are present and context-aware (e.g., 'Path to YAML file or directory containing manifests (required)'). Output schemas are NOT documented in the source code, no return type specifications are visible. Error handling guidance is minimal; tools do not return actionable recovery messages. Tools like kustomize_dryrun, helm_dryrun, flux_dryrun have prerequisites ('Requires select_kube_context to be called first') documented in descriptions, which is good. No security-sensitive parameters are exposed (no API keys, tokens). The dependency on select_kube_context before most tools is clearly signaled but creates a composition concern: agents must call list_kube_contexts → select_kube_context before useful work. This is documented but not enforced programmatically.

• Add explicit output schemas to all tool definitions. Document the return type (object, array, string) and key fields. Example: flux_dryrun should return {status: 'valid'|'invalid', errors: [{message, line}], raw_output: string}. This enables agents to extract validation results and plan error handling.
• Enhance error handling by returning structured error responses with actionable guidance. Example: if select_kube_context has not been called, return {error: 'Context not selected', guidance: 'Call list_kube_contexts to see available contexts, then select_kube_context with your chosen context'}.
• Implement programmatic prerequisites checking. Before flux_dryrun, flux_check, flux_status, kustomize_dryrun, helm_dryrun, and argocd_* tools execute, validate _selected_context is set. If not, return a clear error with the required prerequisite call.
• Add tool annotations (readOnlyHint) to signal read-only tools. Set readOnlyHint: true on list_kube_contexts, flux_check, flux_status, kubeconform_validate, yaml_validate, flux_dryrun, kustomize_dryrun, helm_dryrun, and argocd_* tools. This lets agents safely retry without side effects.
• Expand descriptions for argocd_app_get and argocd_app_diff to 150+ characters. Include when to use each: 'argocd_app_get retrieves the current sync and health status; call this to check if an application is reconciled. argocd_app_diff shows the diff between live and desired state; use this to review what changes are pending.'
• Document the _CONTEXT_SELECT_MIN_DELAY anti-pattern in comments. Consider removing the 2-second delay or replacing it with a prompt/multi-round-trip request (MCP result 'input_required') to explicitly ask the user to confirm context selection before proceeding.