A FastAPI-based MCP server for e-commerce customer service with RAG (Retrieval-Augmented Generation), multi-agent support, knowledge base search, weather integration, and order/product management tools.
Server has 10 tools with basic schemas and descriptions, but significant quality gaps. Most tool descriptions are extremely brief (10-30 chars), well below the 50-200 char LLM-optimized baseline. Parameter descriptions are minimal or absent. No output schemas documented. Error handling is absent, tools return mock data without validation or recovery guidance. Tool naming is acceptable (verb_noun pattern), but composition issues exist: multiple product tools (get_sku_stock, get_product_promotion, get_product_detail) could be unified. Chinese descriptions ('知识库检索', '工具:查天气') are present but lack English equivalents, limiting LLM understanding. No tool annotations (readOnlyHint, destructiveHint) despite clear risk classifications. create_mock_review_task is the only write operation but lacks confirmation or dry-run support.
Check after-sales policy eligibility for an order
Create a mock review task for after-sales requests
Get after-sales order information
工具:查天气
Get logistics tracking information for an order
Get order details by order ID
Get detailed information about a product
Tool descriptions critically short (10-40 chars). LLM-optimized baseline is 50-200 chars. Descriptions like '工具:查天气' and 'Get order details by order ID' lack context for when/why to call the tool, dependencies, or return structure.
No output schemas documented. Tools return dicts (e.g. order_detail returns {order_id, status, shipping_status, pay_time}) but LLMs cannot plan downstream calls without knowing return structure. Breaks tool chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Get promotion information for a product
Get stock information for a product SKU
知识库检索.
Parameter descriptions missing or trivial. 'order_id' param in get_order_detail has description 'The order ID to retrieve details for', does not specify format (UUID? numeric? string?), length, or example. 'order_info' in check_after_sales_policy is object type with no schema or field documentation.
No error handling or recovery guidance. Tools return mock data without validation. If an invalid order_id is passed, no error message guides the LLM to search_orders or validate format. Agents cannot self-correct.
Tool annotations missing. create_mock_review_task is marked WRITE but has no destructiveHint annotation. get_* tools are READ_ONLY but lack readOnlyHint. Agents cannot infer safety/idempotency without annotations.