MCP Server for Code Generation - An intelligent code generator based on Spring AI MCP protocol
This server has critical gaps in definition quality. While 4 tools are registered with basic schemas, the definitions are severely undermined by: (1) tool names that do not follow verb_noun conventions (configureProjectPaths, resetPathConfig are imperative but lack action clarity), (2) descriptions that are overly long (500+ chars) and written in Chinese with critical AI-specific instructions embedded ('【必须首先调用】'), creating confusion about agent vs. human intent, (3) parameters that lack detailed type constraints and validation guidance, (4) no visible error handling specifications or recovery guidance, (5) no documented output schemas for any tool, (6) no pagination or result limits documented despite potential for large dataset returns. The server is currently visible only as a partial source snippet; the full GeneratorTools.java implementation is truncated, meaning input schemas and handler code cannot be fully verified. Per the scoring rules, inferred or partially-visible tool definitions cap at 50; however, the severe quality issues below that threshold warrant a lower overall score.
配置代码生成的目标路径(后端路径、前端路径、SQL输出路径等)。在开始生成代码前,必须先调用此工具询问用户代码存放位置
【必须首先调用】获取业务代码生成的完整流程指南。当用户要求生成业务代码时,AI 必须先调用此工具了解完整流程,然后按步骤执行
重置项目路径配置为默认值
智能扫描项目目录结构,自动推断并推荐前后端代码存放路径。扫描后返回推荐配置,用户确认后可直接使用
Tool names do not follow verb_noun convention. 'configureProjectPaths' and 'resetPathConfig' are imperative but lack clarity. 'getGenerationGuide' and 'scanProjectStructure' are marginally better but could be clearer (e.g., 'get_code_generation_guide', 'scan_project_structure').
Description for 'getGenerationGuide' contains Chinese text and embeds agent instructions ('【必须首先调用】') directly in the tool description. This violates the principle that descriptions should be prompt-engineering-focused, not prescriptive about AI behavior. Descriptions must be neutral and explain WHAT the tool does, not force ordering.
All tool descriptions exceed 150 characters and contain implementation details mixed with user-facing guidance. The baseline for A+ tools is 34 - 392 chars (median ~194). These are between 200 - 300 chars but packed with non-essential context and written in mixed languages.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Parameters in 'configureProjectPaths' (backendRootPath, frontendRootPath, etc.) lack validation constraints. No min/max length specified, no regex pattern documented, no guidance on absolute vs. relative paths. Descriptions are example-heavy ('如: continew-system/src/main/java/top/continew/admin') which LLMs may treat as the only valid format.
No output schemas documented for any tool. The rubric requires that tools document return types so LLMs know what fields to expect. Without visible return schemas, agents cannot plan downstream calls or extract data reliably.
No error handling or recovery guidance documented. If 'configureProjectPaths' receives an invalid path, what does the tool return? If 'scanProjectStructure' cannot find the projectRoot, what are the recovery steps?
Source code is truncated. The full GeneratorTools.java implementation is cut off mid-comment. Cannot verify actual tool implementations, parameter validation, return types, or error handling. Per scoring rules, incomplete visibility of tool definitions caps per-tool scores at 50.