Server has 7 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-50 chars, below ideal 50-200 char range) and input schemas with types. However, descriptions are generic and lack context for LLM selection. Output schemas are undocumented, LLMs cannot predict response structure. No error guidance (e.g., 'User not found. Try search_users()'), no parameter constraints (enums, ranges), and no documentation of write-side effects. Naming is verb-first (good), but descriptions fail to explain WHEN to use each tool or what distinguishes similar tools (e.g., describe_pod vs get_pods). No pagination, no batch operations, and no idempotency guarantees documented.

• Document output schemas for all tools. Example: get_pods returns {namespace: string, pods: [{name, status, node, pod_ip, start_time}]}. Include this in tool docstrings or a separate schema registry.
• Expand descriptions to 50-200 characters and explain WHEN to use each tool. E.g., 'Get all pods in a namespace (use for discovery). For detailed info on a specific pod, use describe_pod().'
• Add parameter constraints: scale_deployment replicas (min=0, max=1000), get_logs tail_lines (min=1, max=10000). State these in parameter descriptions.
• Implement error guidance: catch ApiException, classify as 'not_found', 'permission_denied', 'invalid_input', and return actionable messages with recovery steps.
• Add confirmation pattern for destructive tools: scale_deployment and restart_deployment should support a dry_run=true parameter or return a confirmation token requiring explicit approval.