Scoring was not performed

• Expand tool descriptions to 80-150 characters. Example for list_firewall_rules_cached: 'Retrieve all firewall rules (enabled or disabled) with server-side caching for performance. Pass forceFetch=true to bypass cache and retrieve latest rules from the OPNsense API. Use this instead of get_firewall_rules() when speed is preferred over freshness.'
• Document complete output schemas with example structures. For get_dashboard_data, show: { rules_count: number, vlans: [{vlan: string, tag: string}], cache_hit: boolean, data_timestamp: ISO8601 }
• Add parameter constraints: Define forceFetch and includeDisabled as required booleans with clear defaults. Explain cache TTL behavior (e.g., 'cache is fresh for 5 minutes').
• Distinguish cached tools from uncached variants in tool descriptions, not just names. Explain the performance/freshness trade-off so LLMs know when to select each.
• Add error handling documentation: 'On cache miss, warm_cache may take 10-30 seconds. If timeout occurs, retry with forceFetch=true.' Include guidance for common failure modes.
• Implement output result limits. If dashboard_data includes 500+ rules, cap to 20 and provide pagination via limit/offset parameters. State limits in descriptions.
• Add tool annotations (readOnlyHint) to indicate these tools do not modify state, allowing safer parallel execution.
• Include example successful responses in separate documentation (not descriptions), so LLMs see expected output structure without mistaking examples for constraints.