Loading ToolBench
A knowledge base and document management system with MCP (Model Context Protocol) integration for AI agents. Provides tools for creating knowledge bases, ingesting documents, searching, and retrieving document chunks.
langhuan demonstrates solid tool design with complete input schemas, clear descriptions, and proper risk classification. All 10 tools have documented parameters with types and descriptions. Naming follows verb_noun conventions (knowledge_base_create, document_ingest, knowledge_search). However, descriptions average ~120 chars (below the 194-char baseline for A+ tools) and lack explicit guidance on when to use each tool vs. alternatives. Output schemas are not documented in the source. Error handling is present but generic, no recovery guidance or actionable error messages visible. Tool composition is strong: tools chain well (list → get → search), and parameters accept natural identifiers (knowledge_base_id, document_id). Security is well-handled: no credentials in parameters, destructive operations marked with risk labels.
获取一个 chunk 的完整内容及其修订历史。用于细读检索结果中的某个片段。
删除一篇文档及其所有关联的 chunks 和索引。删除后无法恢复。
获取一篇文档的完整归一化正文（Markdown）与章节大纲（outline 为主线标题路径与行号；无标题结构的文档 outline 为空）。正文默认截断到 50000 字符；细读某个片段所在位置时配合 knowledge_search 的锚点使用。
导入文件或 URL 到知识库。支持两种模式：file 模式（上传文件内容）和 url 模式（将 URL 并入网站源）。
分页列出指定知识库内的全部文档（标题、类型、处理状态）。用于浏览知识库内容概貌；细读某篇文档用 document_get，检索用 knowledge_search。
重新处理一篇处于错误状态的文档。用于在修复问题后重新尝试解析、分块或索引。
Output schemas not documented. Tools return results but LLMs cannot see field structure, forcing them to infer response shape. This breaks downstream tool chaining and increases hallucination risk.
Error handling lacks recovery guidance. No visible error messages that tell LLMs what to do next (e.g., 'Knowledge base not found. Call knowledge_base_list() to see available bases.'). Generic errors block agent self-correction.
Descriptions lack dependency hints and selection guidance. E.g., document_ingest does not explain when to use file vs. url mode, or that knowledge_base_list should be called first to get valid knowledge_base_id values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
查询文档或异步任务的处理状态。支持按 document_id 或 job_id 查询。
创建一个新的知识库，用于组织和管理同类文档。创建后即可用 document_ingest 向其中导入文件，再用 knowledge_search 检索。需指定名称和 Embedding 模型。
列出当前调用方有权访问的全部知识库（API Key 场景为其绑定的知识库集合）。在检索前先调用它了解环境里有哪些知识库，再用 knowledge_base_ids 定向检索。
在一个或多个知识库中进行语义检索，返回相关文档片段及其来源锚点。支持混合检索（向量 + 关键词）。
Destructive operations (document_delete) lack confirmation/dry-run support. No visible pattern to prevent accidental deletion. Agents should be able to preview what will be deleted before committing.
Parameter constraints not fully specified. E.g., knowledge_search.top_k states 'default 10, max 100' but no minimum; chunk_size and chunk_overlap in knowledge_base_create lack range guidance. LLMs may pass invalid values.