MCP server for retrieving and searching Cangjie programming language documentation, supporting multi-version browsing, full-text search, and navigation tree views
Five tools with mostly complete schemas and descriptions. Tool naming is clear and action-oriented (cangjie_docs_overview, cangjie_list_docs, cangjie_search, cangjie_get_doc, cangjie_list_versions). All tools have descriptions between 70-180 characters, meeting the 10-1024 char baseline. Input parameters have type definitions and descriptions. However, output schemas are not documented in the source, only return statements show JSON marshaling without formal schema declarations. Error handling returns generic error strings rather than actionable guidance. Parameter descriptions lack constraint details (e.g., no explanation of what 'confidence' means or valid range for 'level'). No documentation of whether tools are idempotent or have side effects.
获取仓颉语言文档总览,支持 tree(树形导航)/map(文档地图)/overview(分类统计)三种视图
获取仓颉语言指定文档的完整内容,支持按章节获取和元数据过滤
列出仓颉语言文档,支持路径导航(如 std/core)和分类筛选(manual/libs/tools/extra/ohos)
列出仓颉文档的可用版本分支(仅标准 semver 版本,不含 OpenHarmony 等特殊分支),自动识别 latest 和 default
搜索仓颉语言文档,支持多关键词 AND 匹配、分类过滤和相关性排序
Output schemas not documented. Tools return JSON-marshaled responses (map[string]interface{}) but no formal schema is published describing the response structure, field types, or required fields. LLMs cannot plan downstream operations or extract specific data without knowing the response schema.
Error handling provides generic error strings without recovery guidance. For example, 'failed to marshal response: %v' or 'version %s 的搜索引擎未就绪' does not tell the LLM what to do next or whether the error is retryable. Errors should classify as retryable/user-fixable/fatal and suggest alternatives.
Parameter descriptions lack constraint details. 'min_confidence' is described as '最小置信度 (默认0.3)' but does not state valid range (0.0-1.0?) or what confidence means semantically. 'level' is '树形显示深度 (仅navigation/tree视图,默认3,0表示全部)' but minimum/maximum bounds are not explicit.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 11 | - | v1 |
No documented idempotency or side-effect semantics. All tools are marked READ_ONLY, which is correct, but descriptions do not state whether repeated calls with identical parameters return identical results. This is critical for agent retry logic.
Parameter descriptions are in Chinese. While technically valid, descriptions in a non-English language may not be parsed correctly by English-optimized LLMs, reducing effectiveness of the tool discovery and parameter guidance.