InterBase MCP server has 21 tools with consistent naming (verb_noun pattern) and basic descriptions. However, parameter descriptions are minimal or absent for most tools, and output schemas are not documented in the source. Security-sensitive tools (backup, restore, validate) lack confirmation patterns or dry-run support. Error handling guidance is not visible. The server implements role-based access control (VIEW/CRUD/DBA) via Authorization header, which is a positive security pattern, but tool-level permission declarations are missing.

• Document output schemas for all 21 tools. Specify return types, field names, and structure (e.g., open_cursor returns {rows: [{col1: string, col2: int, ...}], row_count: int}). This enables LLMs to chain tools and extract required data.
• Add parameter descriptions to all tools with inputs. For execute_procedure, clarify the structure of 'params' (e.g., 'JSON object with keys matching procedure parameter names, values as strings or numbers'). For schema tools, document what each returns (e.g., get_columns returns array of {name, type, nullable, default}).
• Implement confirmation pattern for destructive operations. Add a 'dry_run' boolean parameter to restore_database and validate_database, or require explicit confirmation via a separate confirm_restore tool. Document this in descriptions.
• Declare tool-level permissions in descriptions. Add text like 'Requires: CRUD or DBA role' to execute_sql, execute_procedure, backup_database, restore_database, validate_database, and sweep_database. This clarifies access control and enables audit logging.
• Add error recovery guidance to tool descriptions. For execute_sql, add: 'On syntax error, check the SQL statement. On permission error, verify your role (VIEW/CRUD/DBA). On timeout, retry with a simpler query.' This guides LLM recovery behavior.