Loading ToolBench
An MCP server that performs Bing searches and fetches webpage content using Playwright browser automation. Supports query splitting via LLM and intelligent search variants.
Two tools with complete JSON schemas and parameter descriptions, but significant gaps in tool-level descriptions and output documentation. bing_search description is in Chinese and lacks English clarity for international LLM agents. fetch_url description is minimal (11 words). Neither tool documents output schema structure. No error handling guidance. Parameters are well-constrained (min/max bounds, defaults) but lack actionable error messages. Tool names follow verb_noun pattern correctly. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite both being read-only operations.
使用 Bing 搜尋並回傳多個網頁的純文字內容。支援子查詢拆分與智慧搜尋變體。
直接抓取指定 URL 的網頁內容並回傳純文字。
Tool descriptions lack English clarity and are too brief. bing_search description is in Chinese; fetch_url is 11 words. LLMs selecting tools need 50-200 character descriptions explaining WHAT, WHEN, and WHY. Current descriptions do not meet the 10-1024 character baseline or provide actionable context.
Output schemas are not documented. LLMs cannot plan downstream calls or extract required fields without knowing response structure. bing_search returns {query, summary, sources, subqueries}; fetch_url returns {url, content}. These must be formally declared in tool definitions.
No error handling guidance. When bing_search returns 'no valid results' or fetch_url fails to load a page, the error message does not tell the LLM what to do next (retry, try different query, ask user). Errors must be categorized as retryable, user-fixable, or fatal with recovery hints.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Tool annotations missing. Both tools are read-only (no side effects), but neither declares readOnlyHint=true. This prevents agents from understanding which tools are safe to call speculatively and which require confirmation. Annotations are part of current MCP spec (2026-07-28).
Parameter descriptions lack format constraints. 'query' and 'url' parameters have minimal guidance. 'url' description says 'needs https://' but does not validate or explain what happens if http:// is passed. Descriptions should state expected format, range, and allowed values explicitly.