Single tool 'run_query' has a clear, concise description (72 chars) and explicit input schema with type and description. However, the output schema is undocumented, the tool returns a formatted string via _format() but LLMs cannot see the structure (columns, rows, masked_columns fields). Parameter descriptions are minimal. No error handling guidance, no enum constraints, no pagination support despite the tool capping results at 50 rows. The server implements strong governance (PII masking, read-only enforcement, audit logging) but does not expose these capabilities through tool metadata or error messages that guide LLM recovery.

• Document output schema in tool description: 'Returns {allowed: bool, columns: [str], rows: [[any]], masked_columns: [str], reasons: [str]}'. Enables LLMs to extract and chain results.
• Add error recovery hints to blocked query responses: 'Query blocked: <reason>. Tip: <recovery action>.' E.g., 'Query blocked: write operations not allowed. Tip: Use SELECT only.'
• Enhance 'sql' parameter description: 'A SELECT-only SQL query. Multi-statement, writes, and file access are blocked. Results capped at 50 rows. PII columns masked unless your role permits.'
• Migrate from STDIO to HTTP+SSE or Streamable HTTP transport to enable remote client access and hosted deployments.
• Add tool annotations (readOnlyHint: true) to signal to clients that run_query is safe to retry and has no side effects.