The server defines 3 tools with basic schemas and descriptions, but multiple critical gaps prevent a higher score. All tools have descriptions and input schemas visible in the code, but descriptions lack depth and actionability. Parameter descriptions are minimal (1-2 words each). No output schemas are documented, only a mention that results are 'CSV format'. Error handling is absent: exceptions are caught and re-thrown as generic RuntimeExceptions with no guidance for LLM recovery. No validation hints, no parameter constraints, no enum values despite opportunities (e.g., query clauses could be constrained). Tool names follow verb_noun pattern (get_*, run_*), which is positive. However, the tools lack composition support, no pagination, no result limiting documented, no IDs returned for chaining. The schema builder code shows proper JSON Schema generation with types, but parameter descriptions are truncated placeholders.

• Document output schema for each tool. E.g., get_tables should declare: 'Returns a CSV with columns: [TABLE_CAT (string), TABLE_SCHEM (string), TABLE_NAME (string)]. First line is header.'
• Example: 'catalog: The database catalog/database name (e.g., "main" in SQLite). Optional; if omitted, uses the default catalog. Required only for multi-catalog systems.'
• Add error recovery hints to tool descriptions. Example for run_query: 'On error, the tool returns the error message. Common errors: invalid column name (call get_columns to verify), nonexistent table (call get_tables to verify), or SQL syntax error (check clause order: FROM, JOIN, WHERE, GROUP BY, ORDER BY, LIMIT).'
• Add limit/offset parameters to get_tables and get_columns. E.g., add optional 'limit' (1 - 1000, default 100) and 'offset' (default 0) parameters. Document in description: 'For large schemas, use limit/offset to page results and avoid context window exhaustion.'
• Add input validation and specific error messages. E.g., if run_query detects UPDATE/INSERT/DELETE, return 'Error: Only SELECT queries are allowed. Attempted: [statement]. Valid clauses: FROM, JOIN, WHERE, GROUP BY, ORDER BY, LIMIT/OFFSET.'
• Return structured data instead of raw CSV. Modify CallToolResult to include parsed JSON with field types and metadata, or at minimum return CSV with a schema header as the first line describing column types.
• Add a constraint hint to run_query's sql parameter: 'Must be a SELECT statement (case-insensitive). Pattern: SELECT ... FROM [table] ... Valid clauses: INNER/LEFT JOIN, WHERE, GROUP BY, ORDER BY, LIMIT/OFFSET. Identifiers quoted with backticks or double-quotes.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry