MCP-SHIELD is an educational security demonstration server, not a production tool. Tool definitions are present but severely deficient: 13 of 15 tools lack proper parameter descriptions (only 'description' field present, no type info visible in schema). Descriptions are present but generic (avg ~80 chars, well below 194-char baseline). No output schemas documented. Critical security tools (exec_shell, dump_env) are intentionally exposed as anti-patterns. Tool naming is inconsistent (scan_host vs scan_host_approved; no verb clarity). Error handling is absent. This server is designed to TEACH vulnerabilities, not to be used in production.

• Document output schemas for all tools. Example: get_balance should return {account_id: string, owner: string, balance: number, currency: string}. Enables LLM planning.
• Add permission gates and confirmation steps to destructive tools (exec_shell, dump_env). Implement a dry_run parameter or require explicit approval before execution.
• Rename scan_host_approved to clarify distinction (e.g., scan_host_with_approval, scan_host_governed). Use consistent verb_noun naming across all tools.
• Expand parameter descriptions with format constraints and ranges. E.g., 'target (hostname or IP; max 255 chars)' instead of 'Target host to scan'.
• Add error handling guidance to tool descriptions. E.g., 'Returns {error: string, retryable: bool, suggestion: string}' for failed calls.