Loading ToolBench
MCP сервер для сбора логов Chrome и отправки их в Cursor
Server has 5 tools with explicit schemas and descriptions. All tools use action verbs (list, connect, get) and have non-empty descriptions (10-50 chars). Input schemas are present with Zod validation and parameter descriptions. However, descriptions are terse (avg 30 chars vs baseline 194), lack WHEN/WHY context, and output schemas are undocumented. Error handling returns text but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Pagination parameters exist on logs/errors but lack min/max constraints. Overall: functional but below production baseline.
Connect to a Chrome tab by title
Get recent Chrome errors
Get Chrome console logs
Get the currently connected Chrome tab
List all available Chrome tabs
Output schemas undocumented. Tools return text content but LLMs cannot infer structure of logs/errors objects (timestamp, message, errorId, stack fields). Downstream tools cannot reliably extract and chain results.
Descriptions too terse (avg 30 chars vs baseline 194). Missing WHEN/WHY context. 'List all available Chrome tabs' does not explain when to call this vs connect-to-tab, or that it requires Chrome DevTools Protocol running. LLMs cannot disambiguate.
Pagination parameters (count, from) lack min/max constraints. No validation that count > 0 or from >= 0. LLMs can pass negative or zero values, causing silent failures or unexpected behavior.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | <=2025-11-25 | v2 |
Error handling returns plain text ('❌ Failed to list tabs: ...') without recovery guidance. LLMs cannot determine if error is retryable, user-fixable, or fatal. No actionable next steps.
No tool annotations. connect-to-tab is destructive (WRITE risk) but lacks destructiveHint. get-chrome-logs is read-only but lacks readOnlyHint. LLMs cannot infer safety/idempotency without explicit hints.