Critical gaps across all dimensions. Tool definitions are inferred from Next.js route handlers rather than explicit MCP registration. No visible input schemas for most tools, only 1 of 5 tools (get_semantic_model) has a documented parameter. Descriptions are present but generic and lack LLM-optimization guidance. No output schemas documented. No error handling patterns visible. The server acts as a Next.js HTTP wrapper around Tableau/Salesforce APIs rather than a proper MCP tool server with formal schema definitions.

• Add explicit MCP tool registration using @mcp/sdk with formal JSON Schema for all 5 tools. Move from inferred route handlers to declarative tool definitions with name, description, inputSchema, and outputSchema.
• Document output schemas for all tools. For get_semantic_model, list the 10 SemanticModelProfile fields with types (e.g., {id: string, name: string, description: string, ...}). For list_* tools, include pagination fields (total_count, next_cursor) and per-item structure.
• Expand descriptions to 50-200 characters. Example: 'list_semantic_models' → 'List all semantic models available in Tableau. Call this first to discover available models before querying metrics or analyzing data. Returns model ID, name, and description.'
• Add input schemas for list_semantic_model_metrics and get_semantic_model_metric. Both accept modelApiNameOrId; get_semantic_model_metric also needs metricNameOrId. Define as required string parameters with descriptions explaining ID vs name resolution.
• Implement error handling with recovery guidance. E.g., if model not found: 'Model "xyz" not found. Call list_semantic_models() to see available models.' If metric not found: 'Metric "abc" not found. Call list_semantic_model_metrics(modelId) to see available metrics.'