Loading ToolBench
Interview MCP Server 是一款强大的MCP Server工具，为公司提供智能化的面试体验。它集成了更加科学实用的简历分析能力，并与主流IDE无缝集成，为面试官提供高效精简的技术面试准备工作流程。
The server defines 3 tools with explicit schemas using Zod, but has significant gaps in LLM optimization and error handling. Tool names are verb-based (good), but descriptions are weak and context-dependent. Parameters are typed with enums, but descriptions lack actionability. No output schemas documented. No error recovery guidance. File I/O operations expose path handling without sanitization documentation.
Evaluate the interviewee's performance based on the interview conversation, Use this tool when mentions /e
When the user mentions to start an interview, use this tool to record the interview conversation. Use this tool when mentioning /record
Generate interview questions based on a PDF resume file. Use this tool when mentions /q
Tool descriptions are minimal and context-dependent. 'Generate interview questions based on a PDF resume file. Use this tool when mentions /q' relies on user mentioning '/q', LLMs cannot infer this trigger pattern. Descriptions must be self-contained and explain WHAT, WHEN, and WHY without external context.
No output schemas documented. Tool implementations call LLM APIs and write files, but the response structure is not formally defined. LLMs cannot plan downstream steps without knowing what fields will be available. Current response is a hardcoded text message, not a structured object.
Parameter descriptions lack actionability. 'Absolute path to the PDF resume file' does not specify format validation, path traversal prevention, or error cases (file not found, unreadable, corrupt PDF). LLMs need explicit format constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
No error handling or recovery guidance. The code throws generic errors ('OPENROUTER_MODEL_ID or OPENROUTER_API_KEY or DEEPSEEK_API_KEY is not set') without telling the LLM what to do next or how to fix it. LLMs receive the error but cannot self-correct.
Path parameters ('absolutePathToPdfFile', 'absolutePathToQuestion', 'absolutePathToConversation') accept arbitrary file paths. No validation shown for path traversal (e.g., '../../../etc/passwd'). File access must be sandboxed to a specific directory or explicitly validated.
API keys are loaded from environment variables but no indication of secret rotation, expiration, or fallback handling. If OPENROUTER fails, the error message exposes the fallback logic (DeepSeek is secondary). LLMs could learn to trigger specific providers.
Resume-to-question writes a markdown file ('resume.md') as a side effect without documenting this in the description. Tool description says 'Generate interview questions' but implementation also creates an intermediate resume file. This hidden state mutates the file system in a way the LLM does not know about.
Record-interview tool description is cryptic: 'use this tool when mentioning /record'. This is a trigger pattern, not a description of what the tool does. The implementation is not shown, cannot verify its full scope or side effects. The tool name 'record-interview' is ambiguous: does it record audio, or log the conversation?