Loading ToolBench
MCP server for scraping Google AI Overviews via a Chrome extension relay
Three tools with basic verb-noun naming (search, follow_up, generate_image). All have descriptions (10-100 chars), but they are terse and lack context for LLM selection. Input schemas are present with types and enums, but parameter descriptions are minimal (5-30 chars). No output schemas documented. No error handling guidance. Tools lack information about prerequisites, return structure, or when to use each variant (fast vs pro mode). Missing idempotency hints and chaining IDs.
Send a follow-up query to continue a conversation thread with Google AI Overview
Generate an image using Google AI's image generation capability
Search Google AI Overviews and get structured markdown response with citations
Output schemas not documented. LLMs cannot infer what fields search() returns (e.g., does it include thread_id, citations, markdown structure?). This breaks tool chaining, follow_up requires thread_id but search's response structure is opaque.
Parameter descriptions are too brief (5-30 chars). 'The search query' and 'The follow-up question' lack context about format, length, or expected content. 'The image generation prompt' does not explain style, size, or constraints.
Tool descriptions lack WHEN/WHY guidance. 'Search Google AI Overviews' does not explain when to use search vs follow_up, what prerequisites exist (Chrome extension running?), or what the response contains. LLMs cannot determine tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If the Chrome extension is not running, or if a query times out, the tools return no actionable error message. LLMs cannot self-correct or retry intelligently.
Mode parameter (fast/pro) lacks explanation. What is the difference in latency, cost, or result quality? When should an agent choose fast vs pro? This ambiguity forces LLMs to guess.