warehouse-mcp demonstrates solid fundamentals: all 16 tools have descriptions (avg ~80 chars, within baseline 34-392), clear verb-noun naming (glossary_lookup, schema_lookup, query, list_schemas, etc.), and visible input schemas with typed parameters. Most tools accept natural identifiers (schema, table, column names) rather than opaque IDs. However, output schemas are not documented in the source code, responses are inferred from tool names and descriptions but not formally specified. Parameter descriptions are present but terse (e.g., 'The schema name' lacks format/constraint details). Error handling is not visible in the tool definitions. No tool annotations (readOnlyHint, destructiveHint) despite all tools being READ_ONLY. Composition is strong: tools chain well (e.g., list_schemas → list_tables → describe_table → query), and each has a single responsibility.

• Document output schemas for all tools. Specify field names, types, and whether results are paginated. E.g., describe_table returns [{name: string, type: string, nullable: boolean, ...}].
• Add readOnlyHint: true to all tool definitions to signal to clients that these tools are safe to cache and retry without side effects.
• Enhance parameter descriptions with explicit constraints: 'schema name (2-64 alphanumeric + underscore)', 'limit (1-1000, default 20)', 'interval (day|week|month|year)'.
• Document error scenarios and recovery paths. E.g., 'If table not found, call list_tables(schema) to see available tables.' Include error codes and actionable messages in tool descriptions.
• Add pagination parameters (limit, offset/cursor) to list_schemas, list_tables, find_columns, and search_value. Document max result limits (e.g., 'Returns up to 100 results; use offset to fetch more.').