Two tools with basic schemas and descriptions, but significant gaps in parameter documentation, error handling guidance, and output schema clarity. execute_sql has a WRITE risk designation but no confirmation/dry-run pattern. set_scripts_dir lacks context on when/why to call it. Neither tool documents return structure or error recovery paths. Descriptions are present but generic (194-char baseline for A+ tools; these are ~120-150 chars). Parameter descriptions exist but lack format/constraint details. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). STDIO transport caps protocol readiness at 50, which cascades to overall quality perception.

• Add tool annotations: mark execute_sql with destructiveHint=true; set_scripts_dir with readOnlyHint=true. Enables LLM to reason about side effects.
• Document execute_sql output schema: 'Returns {rows: [{col1: any, col2: any, ...}], rowCount: number, executedAt: ISO8601}' so LLM knows what fields to extract.
• Expand set_scripts_dir description to ~100 chars: 'Sets the directory where SQL change scripts are stored. Call once at conversation start if you know the path. The UI will watch for new files and enable the Scripts browser. Path must be absolute and readable.'
• Add error recovery guidance to execute_sql: 'If query is rejected, ask the user to clarify intent. If connection fails, retry after 2 seconds. If syntax error, suggest corrections based on error message.'
• Constrain 'query' parameter: add description 'SQL query string (SELECT, INSERT, UPDATE, DELETE). Avoid DROP, TRUNCATE, ALTER. Max 10000 chars.' Prevents hallucinated destructive commands.