The PlanetScale MCP server has 9 tools with solid naming conventions and mostly strong descriptions. Tool names follow the verb_noun pattern (execute_read_query, get_insights, list_cluster_sizes, search_documentation), which aligns with best practices. The execute_read_query tool has an exceptionally detailed description (323 chars) covering what it does, timeout constraints, and RLS caveats. However, critical issues emerge: 7 of 9 tools lack visible input schemas in the provided source code, making it impossible to verify parameter types, constraints, enums, or output documentation. The write_query, get_insights, get_postgres_logs, list_cluster_sizes, payment_methods, query_errors, and query_tags tools show only file paths with no schema definitions. Without visible schemas, those tools cannot score above 30-50 per the hard rules. execute_read_query and search_documentation have documented input schemas with proper types and descriptions, but 7 tools are effectively blind to evaluation. The server lacks per-tool output schema documentation. Error handling is mentioned only in execute_read_query (50-second timeout, RLS warnings); others provide no guidance on recovery or error categorization.

• Provide full input and output schema definitions for all 7 undocumented tools. Document parameter types, required vs optional status, constraints (enums, ranges), and descriptions.
• Add tool descriptions for execute_write_query, get_insights, get_postgres_logs, list_cluster_sizes, payment_methods, query_errors, and query_tags. Each should be 50 - 250 characters, answering: what does it do, when use it vs related tools, what does it return?
• Document output schemas for all tools, listing expected fields, their types, and meanings. For example: 'Returns {success: boolean, query_id: string, rows: array, duration_ms: number}'.
• Add error recovery guidance to each tool: what errors can occur, which are retryable, what should the LLM do next (call search_*, ask user, abort)?
• Implement confirmation/dry-run for execute_write_query. Return a 'confirm' field in the schema; agent must call again with confirm=true to actually execute.
• For list_* and get_*_logs tools, document page/offset/limit parameters and total count in output. State the default limit (suggest 20 - 50) and cap maximum to prevent context window exhaustion.
• Add parameter descriptions for tools missing them. For example, search_documentation should document the 'query' parameter: what search syntax is supported, what does the index cover?
• For query_errors and query_tags, document the from/to time-range rules referenced in insights-tools.ts (e.g., 'Default 24 hours; up to 365 days allowed; ranges >25 hours must be hour-aligned').