Multi-agent customer service system demonstrating MCP integration with Google ADK, featuring inventory and shipping agents backed by FastMCP servers, with security plugins for Model Armor and Gemini Judge
Six tools with basic descriptions and minimal parameter documentation. All tools are READ_ONLY and have simple, single-parameter schemas (or no parameters). Descriptions are present but generic (10-50 chars), falling below the 50-200 char baseline for LLM-optimized descriptions. No parameter descriptions visible in schema definitions. No output schemas documented. Error handling is minimal (generic 'not found' responses). Tools lack actionable recovery guidance.
Check stock level for a product
Get all active (non-delivered) shipments
Get products with stock below threshold
Get return policy information
Get store operating hours
Track a shipment by tracking number
Descriptions are too short (10-50 chars) and lack LLM-optimized context. Baseline is 50-200 chars with WHAT/WHEN/WHY guidance.
Parameter descriptions are missing from schema definitions. 'product_id', 'threshold', 'tracking_number' have no inline descriptions explaining format, constraints, or examples.
Output schemas are not documented. Tools return dicts with fields like 'in_stock', 'products', 'shipments', but LLMs cannot see the structure in advance. This forces runtime discovery and wastes tokens.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | <=2025-11-25 | v2 |
Error responses are generic ('Product not found', 'Tracking number not found'). No recovery guidance. Should suggest next steps: 'Try get_low_stock_products() to see available items' or 'Use get_active_shipments() to list valid tracking numbers.'
Tools 5 and 6 (get_store_hours, get_return_policy) are defined in agent.py as plain Python functions, not registered with FastMCP. Tool definitions are inferred, not explicit.