Server provides 14 tools with basic schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (get_tables, execute_reader, add_row), which is good. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool. Parameter descriptions exist but are minimal (e.g., 'Name of the schema' lacks format/constraint details). No output schemas documented. Error handling and recovery guidance absent. Most tools are READ_ONLY, reducing risk, but write tools (execute_nonquery, add_row, execute_routine) lack confirmation/dry-run patterns. No tool annotations (readOnlyHint, destructiveHint) visible in code.

• Document output schemas for all tools. For get_tables, specify: {type: 'object', properties: {tables: {type: 'array', items: {type: 'object', properties: {name: {type: 'string'}, schema: {type: 'string'}}}}}}. This enables LLMs to plan chained calls.
• Add tool annotations: mark execute_nonquery, add_row, execute_routine with destructiveHint=true; mark all read tools with readOnlyHint=true. This signals to agents which tools have side effects.
• Enhance parameter descriptions with format/constraint details. E.g., 'schema: Name of the schema (case-sensitive, alphanumeric + underscore, max 128 chars)' instead of just 'Name of the schema'.
• Expand tool descriptions to include WHEN to use and what to expect. E.g., 'get_tables: Retrieves all tables in a schema. Call this first to discover available tables before querying columns or data. Returns table names and metadata.'
• Implement confirmation/dry-run for destructive tools. Add optional 'dry_run' parameter to execute_nonquery and execute_routine; return 'Would execute: <SQL>' without committing. For add_row, add 'confirm' parameter requiring explicit true before insert.