Descriptions are generic and under 20 chars for most tools (e.g., 'Return all registered servers' is 31 chars but lacks context on WHEN to call or WHAT to do with results). No parameter descriptions exist. Tools 1-7 have duplicate/overlapping names (get_databases vs get_databases_for_server; get_schemas vs get_schemas_for_database) creating LLM confusion. Tool 9 (resync) has a description copied from tool 8 ('Return all columns for a given table'), clear copy-paste error. No output schemas documented. No error handling guidance. No pagination support despite metadata discovery tools that could return large result sets.

• Consolidate duplicate tools: remove get_databases, get_schemas, get_tables (parameterless variants). Keep only the _for_* versions with required ID parameters. This eliminates LLM confusion and reduces tool count from 9 to 6.
• Add complete input schemas to all tools. E.g., get_databases_for_server should have {"type": "object", "properties": {"server_id": {"type": "integer", "description": "The ID of the server (required)"}}, "required": ["server_id"]}.
• Document output schemas for all tools. E.g., get_databases_for_server returns {"databases": [{"id": integer, "name": string, "size_mb": integer}], "total": integer}. Include all fields needed for downstream tool calls.
• Add pagination: limit (1-100, default 20) and offset (default 0) parameters to discovery tools. Return total_count in responses so LLMs know if more results exist.
• Rewrite tool descriptions to be LLM-optimized (50-200 chars). E.g., 'List all SQL Server instances registered in this MCP server. Call this first to discover available servers before querying databases.' Include WHEN to call and WHAT to do next.