Three tools with minimal descriptions (10-30 chars, well below 50-200 char baseline). No output schemas documented. execute_sql is destructive but lacks confirmation/dry-run pattern. Tool names follow verb_noun convention (positive), but descriptions are too terse to guide LLM selection. No error handling guidance visible. Schema registration uses mcp-go framework correctly, but parameter descriptions are absent.

• Expand tool descriptions to 50-200 chars. Example: 'get_tables' → 'Retrieve all table names and schemas in the connected database. Call this first to discover available tables before querying specific ones.'
• Add parameter descriptions with format/constraint hints. Example: table_name → 'The name of the table to inspect (case-sensitive, must exist in current database).' query → 'SQL statement to execute (SELECT, INSERT, UPDATE, DELETE, or DDL). Queries run with current connection permissions.'
• Document output schemas for each tool. Example: get_tables returns [{name: string, row_count: number, columns: [{name: string, type: string}]}].
• Add dry-run support to execute_sql: new parameter 'dry_run: boolean' (default false) that validates syntax without executing. Return 'Query validated. Execution would affect X rows.' before actual execution.
• Enhance error messages with recovery guidance. Example: 'Table "users" not found. Available tables: [users_archive, user_profiles, user_settings]. Did you mean one of these?' instead of 'failed to get table detail'.