Single tool 'execute_sql' has a well-structured description (194 chars) and clear parameter documentation. Input schema is complete with proper types (string) and descriptions for both 'query' and 'warehouse_id' parameters. Output schema is documented in the description ('status, warehouse_id, columns, rows, row_count'). Error handling returns structured error objects. However, the tool lacks formal output schema definition in the registration, parameter constraints (no enum for warehouse_id fallback behavior), and does not declare idempotency or destructiveness hints. The description is good but could be more concise for LLM parsing. No pagination, batch operations, or recovery guidance for common failure modes (e.g., warehouse not found, query timeout).

• Add formal output schema to the tool registration: `dict[str, Any]` with typed fields (status: str, warehouse_id: str, columns: list[str], rows: list[list[Any]], row_count: int, error?: str).
• Declare tool annotations: add `readOnlyHint=True` to the @mcp.tool decorator to signal to clients that this tool does not modify state.
• Document warehouse_id format in the parameter description: 'Databricks warehouse ID (UUID format, e.g. abc123def456). If omitted, uses DATABRICKS_WAREHOUSE_ID env var.'
• Enhance error messages with recovery guidance: distinguish timeout errors ('Query exceeded 30s timeout. Try simplifying the query or increasing wait_timeout.') from invalid warehouse errors ('Warehouse abc123 not found. Check DATABRICKS_WAREHOUSE_ID or pass a valid warehouse_id.').
• Add result limit documentation: 'Returns up to 10,000 rows. For larger datasets, use LIMIT clause in the query or add offset/limit parameters to this tool.'

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained