Loading ToolBench
MCP server for Trawlia: web search and extraction for AI agents.
Two well-named tools with comprehensive input schemas and strong descriptions. trawlia_search and trawlia_extract both follow verb_noun naming and include detailed parameter documentation. Descriptions are 150-250 chars, within the 10-1024 baseline. All parameters have types and descriptions. However, output schemas are not documented, LLMs cannot see what fields to expect in responses. Error handling is excellent (readable, actionable messages), but no output structure guidance limits downstream tool chaining. No tool annotations (readOnlyHint, idempotentHint) despite both being read-only operations.
Fetch URLs you already have and return their readable content, with navigation, cookie banners and newsletter modals removed. Costs 1 credit per 5 URLs, so pass the whole list at once rather than calling this repeatedly. Use it when you already know which pages you want; use trawlia_search when you do not.
Search the live web and get back ranked, deduplicated results. Use search_depth 'basic' (1 credit, ~10x faster) for most questions: it returns titles, URLs and snippets, which is enough to answer or to decide what to read. Use 'advanced' (2 credits) only when you need the actual text of the pages, because it fetches and extracts every result. Set topic 'news' for anything time-sensitive; it is tuned for recent events and returns publication dates.
Output schemas not documented. LLMs cannot see what fields (title, url, snippet, publication_date, etc.) are returned by trawlia_search or trawlia_extract. This forces agents to guess field names and breaks downstream tool chaining.
No tool annotations. Both tools are read-only (search and extract do not modify state), but lack readOnlyHint annotations. This prevents clients from optimizing caching, retry logic, or safety checks.
max_results parameter lacks upper bound constraint in schema. While description says 'typically 20 to 60', no maxItems is enforced, allowing LLMs to request thousands of results and exhaust context.
No pagination guidance in output. trawlia_search returns 'typically 20 to 60' results but does not document how to fetch additional pages or whether a next_cursor/total_count is provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Error messages reference 'request_id' but tool descriptions do not explain what to do with it or how to report issues. Incomplete recovery guidance.