8 tools with clear verb-noun naming (dns_list, dns_add, dns_update, dns_delete, zone_list, cache_purge, firewall_list, firewall_add). All have descriptions (avg ~50 chars, baseline 194). Input schemas visible with types and descriptions for all parameters. However, descriptions are terse (10-50 chars), below the 50-200 char LLM-optimized range. No output schemas documented. No error handling guidance. No enums for constrained inputs (e.g., 'mode' in firewall_add accepts 5 values but no enum). Token parameter exposed across all tools (security risk). No pagination support for list tools. Missing idempotency/destructive hints.

• Remove 'token' parameter from all tools. Implement server-side secret injection: read CF_API_TOKEN from environment at startup, pass to resolveClient() without exposing to LLM. Document in server README that users must set CF_API_TOKEN env var.
• Expand all tool descriptions to 100-200 chars. Include WHAT (action), WHEN (use case), and WHAT IT RETURNS. Example: 'dns_list: Retrieves all DNS records for a Cloudflare zone. Use to audit current DNS configuration or find a record before updating. Returns array of records with id, type, name, content, TTL, and proxy status.'
• Add enum constraints to 'type' (dns_add, dns_update) and 'mode' (firewall_add) parameters. Use JSON Schema enum field: {"type": "string", "enum": ["A", "AAAA", "CNAME", "MX", "TXT"]}. This prevents hallucinated values and self-documents valid options.
• Add pagination to list tools: add 'limit' (default 20, max 100) and 'offset' (default 0) parameters. Return {items: [...], total: N, offset: 0, limit: 20} so agents can iterate. Document in descriptions: 'Returns up to 20 records per call; use offset to paginate.'
• Document output schemas in tool descriptions or a separate schema file. For each tool, specify returned fields, types, and examples. E.g., dns_list returns [{id: string, type: string, name: string, content: string, ttl: number, proxied: boolean, created_on: string}].