基于同程旅行网API接口二次封装的MCP Server服务,目前支持火车票查询和系统时间查询功能
This server has 2 tools with basic schemas and descriptions, but significant quality gaps prevent it from achieving a higher score. Tool names follow verb_noun convention (query_, get_), which is good. However, descriptions lack LLM-optimization guidance, parameter descriptions are present but minimal, output schemas are entirely undocumented, and error handling provides generic responses without recovery guidance. The server makes actual HTTP calls to a third-party API (同程旅行, Tongcheng Travel) but exposes no pagination, rate-limiting, or API-specific error context to the caller. Tool 1 (query_train_tickets_list) returns unstructured JSON with field name transformations (Chinese labels) that would confuse LLMs trying to chain calls. Tool 2 (get_current_system_time) is trivial and has no parameters. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. Overall, these are functional tools but fall well short of production quality.
查询当前系统时间工具,它可以帮助你获取当前系统日期时间,返回格式为:yyyy-MM-dd hh:mm:ss,可用于火车票查询等需要时间验证的场景。
火车票车次列表查询功能,查询指定日期、出发地和目的地之间的火车票信息。输入用户提供的出发站出发城市、到达站到达城市、出发日期进行查询结果,返回包含车次列表、出发/到达站信息、座位类型及票价等详细信息。本工具提供中国境内火车票班次信息的全面查询服务,数据实时对接同程旅行网官方接口。 提供不同座位类型(商务座/一等座/二等座等)的实时票价
Output schemas are completely undocumented. LLMs cannot predict the structure of tool responses, making downstream chaining and field extraction error-prone. The query_train_tickets_list response is visible in code (convertFieldNames function transforms API data into Chinese-labeled fields like '用户查询出发站', '车次列表'), but this is never declared to the caller.
Error responses provide no recovery guidance. When the API call fails (e.g., 'Invalid station name' or network timeout), the tool returns a generic error message like '获取火车票城市编码数据失败' (Failed to get train ticket city code data) with no suggestion of what the LLM should do next, retry? Call a different tool? Ask the user?
Parameter descriptions are minimal and lack format/constraint details. The depDate parameter states 'format must be yyyy-MM-dd' in the description, which is good, but there is no guidance on valid date ranges (e.g., must be today or future, max 60 days ahead). The depStationName and arrStationName descriptions mention 'support station name or city name' but provide no examples of invalid input or how the tool handles partial matches.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Tool descriptions are verbose (Chinese text > 200 chars) and lack LLM-optimized structure. They state what the tool does and mention 'real-time data from Tongcheng Travel API', but do not clearly answer: When should the LLM choose this tool over alternatives? What are the prerequisites? The description reads as a marketing copy, not an agent prompt.
No pagination, limit, or result count constraints. The tool can return an unbounded list of train results in a single response. If there are 50+ trains on a popular route, the response will be massive, wasting tokens and degrading LLM reasoning. No 'limit' or 'page' parameter is offered.
Response field names are in Chinese (用户查询出发站, 车次列表, 出发火车站是否是经停站). While this matches the intended user language, it makes tool chaining and downstream LLM reasoning difficult if the system prompt is in English or if the response is passed to another tool expecting English field names. Consider returning both English and Chinese labels, or making this configurable.
No input validation or constraint declaration for required parameters. The code checks if depStationName, arrStationName, and depDate are empty strings, but does not validate date format (yyyy-MM-dd), allow future dates only, reject dates > 60 days ahead, or handle station name ambiguities. The description mentions 'support station full name (e.g. Beijing West Station) or city name (e.g. Beijing)' but the tool does not document what happens if the user provides an ambiguous name (e.g., 'Beijing' could match 5 stations).
Tool descriptions do not clarify when to use query_train_tickets_list vs get_current_system_time, or if they work together. The description of get_current_system_time mentions it is 'useful for train ticket search scenarios requiring time validation', but query_train_tickets_list does not explain why an LLM would need to call get_current_system_time first (presumably to validate that depDate is not in the past).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools are read-only and idempotent (calling them multiple times with the same input returns the same result), but these properties are not declared. Tool registration does not include risk or annotation metadata.