Critical failures across all dimensions. All 9 tools use identical, overly generic schemas with a single 'input' string parameter lacking any structure or validation. Descriptions are minimal (12-26 chars, well below the 50-200 char baseline for A+ tools). No parameter descriptions exist. No output schemas documented. No error handling guidance. Tools are thin pass-throughs to an external API with no validation, composition, or agent-friendly design. This server treats MCP as a simple HTTP proxy rather than a proper tool interface.

• Replace generic 'input' string parameter with structured schemas. For supplier_radar, define: {supplier_name: string (required), region: enum[China, Vietnam, India], min_rating: number 0-5, limit: number 1-100}. Each tool needs domain-specific parameters.
• Expand descriptions to 80-150 chars explaining WHAT, WHEN, and WHAT IT RETURNS. Example: 'Search Chinese suppliers by name, region, and rating. Returns supplier profiles with contact info, certifications, and audit history. Use before supplier_evaluator to get candidate list.'
• Add output schema documentation in tool descriptions or as a separate field. Example: 'Returns {suppliers: [{id, name, region, rating, certifications, contact_email}], total_count, next_cursor}'.
• Implement error classification. Wrap callBankr() to return {status, error, retryable: boolean, suggestion: string}. Example: 'Supplier not found. Try search with partial name or different region.'
• Rename tools to verb_noun pattern: supplier_radar → search_suppliers, spare_parts → list_spare_parts, warranty_analyzer → analyze_warranty_claim. Add action verbs so LLMs parse intent from names.