Single tool with critical definition gaps. Tool name 'compare_visual_specs' lacks action verb clarity (compare is weak; should be 'analyze_visual_specs' or 'get_visual_spec_comparison'). Description is 95 chars, acceptable length but vague on WHEN to use it and what it returns. Input schema visible but parameters lack descriptions entirely (product_a and product_b have descriptions in the schema, but they are generic defaults, not actionable constraints). No output schema documented, LLM cannot predict response structure. No error handling, no validation, no recovery guidance. Tool definition is hardcoded in mcp_server.py with no explicit schema registration visible in tools/list response.

• Rename tool to 'analyze_visual_spec_comparison' or 'get_visual_specs_comparison' to start with a clear action verb.
• Add parameter descriptions: 'product_a: Name or model ID of the first product (e.g., "Unitree G1 Humanoid", "Boston Dynamics Atlas"). If not found, returns available products.' Similar for product_b.
• Document output schema in tool description: 'Returns {comparison_id, compared_products, spec_delta_matrix: [{dimension, prod_a, prod_b, advantage}], visual_recommendation}'.
• Implement actual parameter handling in client.compare_visual_specs(), currently ignores product_a and product_b arguments and always compares the same two robots.
• Add error handling: if product not found, return {error: 'Product not found', available_products: [...], suggestion: 'Did you mean...?'} instead of silent fallback.