The server provides 16 tools with generally clear naming conventions (verb-noun patterns like show_*, describe_*, list_*) and documented input schemas. However, several critical gaps reduce the overall quality: (1) Tool descriptions are short but mostly adequate (10-150 chars); (2) Output schemas are completely undocumented, no tool specifies what fields or structure it returns, forcing LLMs to infer; (3) Error handling guidance is absent across all tools; (4) Several tools lack parameter-level descriptions (e.g., show_catalogs has no parameters, so description adequacy is N/A); (5) Parameter descriptions are present but minimal (e.g., 'The name of the catalog' is bare, lacks format/constraint guidance); (6) Tool composition is sound, tools chain naturally (list catalogs → show schemas → show tables); (7) The execute_query tool is dangerously underdescribed given its WRITE risk; (8) Destructive operations (expire_snapshots) lack dry-run or confirmation patterns. Per-tool analysis shows 13/16 tools score 60-70 (fair), with show_catalogs and show_catalog_tree scoring lower due to no parameters, and execute_query scoring lower due to insufficient risk/constraint documentation.

• Document return type for every tool. Specify: 'Returns: str (JSON-formatted list of catalogs, each with name, description, and count of schemas)' or similar. Use JSON Schema or TypeScript-like notation for clarity.
• Add output schema examples to tool descriptions. E.g., 'Returns: {catalogs: [{name: string, description?: string, schemas_count: integer}]}'
• Add error recovery guidance to every tool, especially execute_query. E.g., 'If the query times out (>30s), simplify the filter or call show_schemas to confirm the table exists first.'
• Implement dry-run for expire_snapshots: add a 'dry_run' boolean parameter (default=true) to show what would be deleted before committing.
• Enforce parameter constraints: convert retention_threshold example into an enum or regex. Document: 'Must match pattern: ^\d+[dhm]$ (e.g., 7d, 30d, 24h). Default: 7d'
• Add pagination to discovery tools: add 'limit' (1 - 1000, default=100) and 'offset' or 'cursor' parameters to show_catalog_tree, show_query_history, describe_table.
• Upgrade execute_query description: 'Executes arbitrary SQL. Returns results as JSON. WARNING: modifying queries are not supported, use separate tools for optimize/expire_snapshots. Max runtime: 5 minutes. If query exceeds this, call execute_query with a narrower filter or reduced dataset.'
• Add permission hints to tool descriptions. E.g., 'Requires Trino EXECUTE privilege on target catalog and schema.'
• Document chaining IDs: if describe_table returns a table_id, ensure show_table_properties and other downstream tools accept that ID. Otherwise, specify in the response what ID format is needed.

Inferred effective spec: 2026-07-28+.

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