Loading ToolBench
An AI agent that breaks down complex queries into multi-step actions using MCP tools for web search, file/image upload, and markdown document generation
This MCP server has critical gaps in tool definition quality. Of the 4 tools examined, all lack comprehensive input schema documentation, and most have minimal descriptions. Tool definitions appear inferred from code rather than explicitly registered with full schemas. The server uses Chinese descriptions, which while not inherently problematic, suggests incomplete internationalization for LLM consumption. No visible output schema documentation, error handling guidance, or parameter validation. The server implements basic tool functionality but falls far short of production-grade definition standards.
上传本地文件
上传本地图片
将文本以markdown形式保存在本地
搜索互联网内容
file_upload and image_upload tools have ZERO input schema definition and NO description text. Per HARD SCORING RULE, these must score 0 for schema and description respectively.
All tool descriptions are extremely brief (10-30 chars) and in Chinese. The descriptions lack WHAT/WHEN/WHY guidance required for LLM selection. 'Upload local file' does not explain when to use this tool, what formats it accepts, size limits, or what it returns.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., does web_search return raw HTML, structured results with URLs and titles, or plain text? Unknown.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 32 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
file_upload and image_upload have empty input schemas ({}). This violates the pattern that every parameter must have a description and type. LLMs cannot infer what these tools require.
No error handling guidance. Tools do not document what errors they can return, whether they are retryable, or what the LLM should do next. E.g., what if a web search times out? What if file_upload rejects the format?
No permission checks or audit trails visible. Tools like markdown_generate (marked WRITE) and file_upload should declare scope requirements and log execution for compliance.
Tool naming inconsistency: 'file_upload' and 'image_upload' do not clarify WHERE files go or what the return value is. Are they uploading to cloud storage, a local directory, or a database? This ambiguity forces LLMs to guess.
markdown_generate parameter 'doc_name' lacks format constraints. No mention of file extension, length limits, or character restrictions. LLMs may pass invalid names like 'my file!@#.md'.