Loading ToolBench
Web search and content extraction via OrioSearch API. Exposes search, extract, and streaming search tools for LLM clients (Claude Desktop, Cursor, etc.) using SSE transport.
OrioSearch exposes 3 tools with complete JSON schemas and reasonable descriptions. All tools have verb-noun naming (search, extract, search_stream) and parameter types are declared. However, descriptions are generic and lack LLM-optimized guidance on when to use each tool vs. alternatives. Parameter descriptions are present but minimal (e.g., 'Search query.' vs. 'Enter a search query to find web results; use specific keywords for better relevance'). Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent from tool definitions. The extract tool lacks guidance on format choice (markdown vs. text). search and search_stream are near-duplicates with unclear differentiation.
Extract page content from URLs.
Search the web. Returns results with title, URL, snippet, and relevance score.
Search the web with parallel content extraction. Faster than search when using search_depth="advanced" or include_answer=True.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Responses are formatted as markdown text in _format_search_response() but this is not exposed to the MCP client.
search and search_stream are near-duplicates with identical parameters. Descriptions do not explain when to use search_stream over search (only a hint in search_stream description: 'Faster than search when using search_depth="advanced"'). LLMs will struggle to choose between them.
Parameter descriptions are minimal and lack actionable guidance. E.g., 'Search query.' (9 chars) vs. expected 50-200 chars. 'Regex to filter results by title/URL/content.' lacks examples of valid regex patterns. No guidance on when to use include_answer, include_raw_content, or include_images.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
No error handling guidance in tool definitions. If a search fails or returns no results, LLMs have no recovery path. No documentation of retryable vs. fatal errors.
extract tool format parameter (markdown vs. text) lacks guidance on when to choose each. No description of output differences or use cases.