Loading ToolBench
A job search and crawling tool built with Model Context Protocol
The MCP Jobs server has critical definition quality gaps that severely limit its production readiness. Both tools have minimal descriptions in Chinese, lack comprehensive parameter documentation, and schemas are present but incomplete. The server implements only 2 tools focused on job searching via web crawling. Naming follows verb-noun conventions (mcp_search_job, mcp_job_detail), but descriptions are insufficient for LLM-driven tool selection. Parameter schemas exist but descriptions are sparse and in Chinese, creating a language barrier. No output schema documentation is provided, forcing LLMs to infer result structure. Error handling is absent from the visible tool definitions. The codebase shows a web crawler implementation using Playwright, but the MCP tool wrapper does not expose sufficient metadata for reliable agent interaction.
获取职位详情信息，包括职位名称、公司名称、薪资范围、工作地点、发布时间等。
搜索职位信息，包括职位名称、公司名称、薪资范围、工作地点、发布时间等。
Tool descriptions are minimal (under 20 characters meaningful English content) and in Chinese only. 'mcp_search_job' description is 80 chars but primarily in Chinese; 'mcp_job_detail' is similarly opaque to English-speaking agents.
No parameter descriptions provided. The 'keyword' parameter in mcp_search_job has no description field in the schema. Required parameters like 'url' in mcp_job_detail lack context explaining format, validation rules, or examples.
Output schema not documented. Neither tool specifies what structure is returned (object shape, field types, pagination). LLMs cannot plan downstream operations or extract required data.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 41 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 21 | - | v1 |
No pagination support visible. mcp_search_job accepts a 'page' parameter but no documentation of page size limits, total count, or next_cursor. Results could be unbounded. Per pattern, tools returning lists must accept limit/offset and return total count or pagination token.
Enum constraints missing. Parameters like 'salary' and 'workYear' accept free-form strings with no enum or validation. Agents will hallucinate invalid values.
No error handling guidance. Tool definitions do not specify error cases (invalid URL, network timeout, no results) or recovery steps. Per pattern, errors must tell the LLM what to do next.
Parameter 'url' in mcp_job_detail provides no validation hints. No description of expected URL format, domain restrictions, or parameter requirements (is it a full URL? Can it be a relative path?).
Composition clarity lacking. The relationship between mcp_search_job (returns list) and mcp_job_detail (accepts URL) is implicit. No documentation of whether search results include URLs for detail fetching, or how to chain the tools. Per pattern, tool A's output must contain IDs/references tool B needs.
Language barrier: descriptions and parameter names in Chinese create friction for English-speaking LLMs and developers. All tool and parameter documentation should be in English for international adoption.