MCP server that provides tools to interact with MeterSphere API, enabling AI to access API lists, API details, and user information from MeterSphere testing platform
This server exposes 3 tools via Spring AI annotations with basic parameter schemas. However, significant definition quality issues prevent a higher score: (1) Tool descriptions lack depth and fail to explain WHEN to use each tool vs alternatives, contradicting pattern:tool-description guidance. (2) Parameter descriptions are minimal or absent in the visible source (e.g., 'protocols' array in getApiList has no value constraints or explanation of what 'HTTP', 'TCP' mean). (3) Output schemas are completely undocumented, there is no evidence of what fields getApiList, getApiDetail, or getUser return, preventing LLMs from planning downstream calls. (4) The code shows tools are 'marked as READ_ONLY' but no error handling, rate-limiting, or recovery guidance is visible. (5) Tool names are reasonable (verb_noun pattern), but composition is unclear, why separate getApiList and getApiDetail instead of one parameterized tool? The visible code fragment is incomplete; full tool implementation files are not provided, making it impossible to verify parameter validation, response shaping, or error messages.
API详情获取工具,通过API ID获取单个API的详细信息
API列表获取工具,获取项目下的API列表信息,支持分页、排序、关键词搜索和协议筛选
用户基本信息获取工具,获取用户相关信息,包括权限和角色,所属组织和项目等
Output schemas completely undocumented. No evidence of what fields getApiList, getApiDetail, getUser return. LLMs cannot plan chained calls or extract required data. Violates pattern:response-shaper and pattern:tool.
Tool descriptions are generic and lack WHEN context. They explain WHAT the tool does but not WHEN to use it instead of a similar tool, or what prerequisites exist. Violates pattern:tool-description guidance that descriptions must include WHAT, WHEN, and prerequisites.
Parameter constraints missing. 'protocols' array in getApiList has no enum declaration or explanation of valid values. 'apiId' lacks type constraint (UUID? string pattern?). 'current' and 'pageSize' have no min/max bounds, allowing absurd values that could hang APIs. Violates pattern:constrained-input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 31 | - | v1 |
getUser has zero input parameters (empty {}). This is highly ambiguous, does it fetch the current authenticated user? A specific user? How would an LLM know? If it fetches current user, this must be explicitly stated. If it accepts user_id or user_name, parameters are missing. Violates pattern:tool-description and creates ambiguity.
No error handling, recovery guidance, or validation visible. No evidence of actionable error messages per pattern:recovery-guide. If an API call fails, LLMs have no guidance on whether to retry, which tool to call next, or what went wrong. The code shows HTTP client calls but no visible error wrapping.
No pagination limit enforcement visible. getApiList accepts 'pageSize' with no documented upper bound. Returning thousands of APIs could exhaust context window. Per pattern:paginated-result, tools should cap results and enforce reasonable defaults.
Credentials (meterSphereAddress, accessKey, secretKey) are stored in ServerConfig. If these appear as tool parameters in actual implementation, this violates pattern:secret-injection fundamentally, secrets must never be tool parameters. Code shows they are stored server-side, which is correct, but full parameter handling is not visible.
Tool composition unclear. getApiList and getApiDetail are separate tools. Is there a reason? If an LLM almost always needs details after listing, combining them (or having getApiList return minimal detail IDs + summaries) would be more efficient per pattern:tool-composition.