Loading ToolBench
MCP server for Synorive — let Claude Code search your local multimodal library (documents, code, images, video) and fact-check claims against the open web. 24 tools, fully offline-capable.
Synorive MCP defines 7 tools with Chinese-language descriptions and reasonable parameter schemas. All tools have descriptions (10-200+ chars) and input schemas with types. However, descriptions lack English translations and actionable guidance for LLM selection. Parameter descriptions are present but terse. Output schemas are not documented. Error handling and recovery guidance are absent. Tool naming follows verb_noun convention (synorive_search, synorive_ingest) but descriptions are optimized for human users, not LLM reasoning. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.
分析单个文件并同步等结果返回（和 synorive_ingest 的区别是这个会等完成）。适合"我想知道这个文件里有什么"的场景。大文件会比较慢。
按 id 取一条内容的完整正文。id 从 synorive_search 的结果里拿。想读一个文件的具体内容时用这个，比让用户复制粘贴省事。
看库里自动抽取的实体（邮箱、链接、金额、日期、版本号、机构名）以及它们的共现关系。用来顺藤摸瓜：从一个实体找出所有提到它的内容。
把本机的文件或整个目录加入索引，之后就能搜到。支持文档、代码、图片、视频、音频、网页存档。这是后台任务，会立刻返回一个 jobId，用 synorive_status 看进度。重复投喂同一份内容会自动跳过（按内容指纹去重），不会重复占空间。
在用户的本地内容库里搜索（文档、代码、图片里的文字、网页存档、视频台词）。支持中文语义检索——描述内容也能搜到，不用记文件名。查询串里可以直接写筛选指令：type:pdf、date:>2026-01、date:最近7天、size:>10mb、in:D:\项目、tag:重要、src:link、-排除词、"精确短语"。论文可以按章节过滤：section:方法 / section:结果 / section:结论（中英文都认，按子串匹配所以「3.2 Experimental Method」这种带编号的标题也能命中）。视频和音频的结果会带上时间点，可以直接跳到那一秒。
给一条内容的 id，找库里和它相似的其它内容。如果那条是图片，还能找出视频里相似的**镜头**并给出时间点。
Descriptions are in Chinese only; LLMs trained primarily on English will struggle to understand when/why to invoke tools. No English translations provided.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., itemId from search results for get_content). Responses are unstructured.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). synorive_ingest is clearly destructive (WRITE risk) but unmarked; synorive_search is read-only but unmarked. LLMs cannot infer safety properties.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
按时间桶统计库里的内容分布。回答"我那段时间在忙什么""某个月存了些什么"这类问题。
Parameter descriptions are minimal (5-20 chars). 'query' in synorive_search says '查询词，可含筛选指令' but does not explain in English what filtering directives are valid or how to use them. LLMs cannot self-correct invalid queries.
No error handling or recovery guidance. If synorive_ingest fails (e.g., file not found, unsupported format), no actionable error message or next-step guidance is documented.