Loading ToolBench
This server provides 10 search and crawl tools with basic descriptions and input schemas. However, it suffers from significant quality gaps: parameter descriptions are minimal (most lack detail on format, range, or constraints), output schemas are not documented, and error handling guidance is absent. Tool names follow verb_noun convention (positive), but descriptions are generic and often lack WHEN/WHY context. The server prioritizes breadth (10 tools across multiple search APIs) over depth of documentation. No tool annotations, no enumerated parameters, no pagination guidance. Descriptions average ~80-100 chars, below the A-grade baseline of 50-200 chars with full actionability.
Search arXiv for academic papers with title, summary, and URL results
Search Baidu search engine and return URL results
Search Bing search engine and return results with titles, snippets, and links
Perform automated browser actions using the browser-use package. Enables tasks such as navigating to URLs, interacting with web elements (clicking buttons, entering text), and extracting data from web pages using natural language task descriptions.
Crawl a web page and extract markdown content
Crawl a web page using proxy configuration and extract markdown content
Search DuckDuckGo and return URL results with priority ranking
Output schemas not documented. No indication of what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls or extract the right data.
Parameter descriptions are minimal and lack actionable constraints. 'max_results' appears on 7 tools but has no documented range (min/max), no guidance on behavior with out-of-range values, no default documented. 'query' parameters lack format or length constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Search Google using Custom Search API and return URL results with priority ranking
Search using Tavily search API and return comprehensive results
Search Wikipedia and return results
Tool descriptions are generic and generic (45-55 chars average). They do not answer WHEN to use each search tool instead of the others. LLMs cannot disambiguate between 6 similar search tools without explicit guidance (e.g., 'Use for academic papers', 'Use for Chinese results', 'Use for comprehensive web results').
No error handling guidance. Tools do not document what to do if search fails, if no results are found, or if API credentials are missing/invalid. No recovery instructions for agents.
No pagination support documented. Search tools that return lists do not indicate max_results behavior, whether results are ranked, or how to retrieve next page of results. Returning all results risks context window explosion.
Tool naming confusion: 'search_tavily_allInfo' violates verb_noun convention (contains 'all' and 'Info' which are unclear). Should be 'search_tavily' or 'tavily_search'.
Proxy configuration for crawl_page_via_proxy is not documented. No description of how to configure proxies, what format proxy_url should take, or what happens on proxy failure.
No tool annotations present. No readOnlyHint, destructiveHint, or idempotentHint fields. All tools are marked READ_ONLY in metadata but this is not surfaced in the tool definition itself.