Single tool with basic schema and description. Tool name 'extract-web-data' uses verb-noun pattern (good), but description lacks critical context about prerequisites, rate limits, or failure modes. Input schema is present with proper types and descriptions for both parameters (url, prompt). However, no output schema is documented, LLMs cannot predict the structure of returned data. No error handling guidance provided. Tool is read-only (safe), but lacks idempotent/destructive annotations. Baseline: 194 chars avg description; this one is 155 chars (acceptable). Both parameters have descriptions (good). Missing: output schema documentation, error recovery guidance, rate limit warnings.

• Add explicit output schema to tool definition: document that result contains a 'data' field with object type, and describe expected top-level keys (e.g., 'products', 'articles', 'metadata'). Example: 'Returns {data: {[key: string]: any}} where data structure depends on the extraction prompt.'
• Expand description to include: 'Requires AGENTQL_API_KEY environment variable. Supports public URLs only. Extraction may take 5-30 seconds. Returns structured JSON matching the natural language prompt. If extraction fails, returns error with reason (timeout, invalid URL, extraction failed).'
• Add readOnlyHint: true and idempotentHint: true to tool definition to signal safety to agents.
• Implement structured error responses: instead of generic 'AgentQL API error', return {error: 'extraction_failed', reason: 'timeout|invalid_url|api_error', suggestion: 'Retry with a simpler prompt or verify URL is publicly accessible'}.
• Document result limits in description: 'Results are capped at 50KB of JSON. For larger extractions, break the prompt into multiple smaller queries.'