Single tool 'render' has excellent schema completeness with comprehensive parameter documentation, proper type constraints (enums, min/max bounds), and structured output schema. Descriptions are detailed and actionable (194 chars for tool, 50-150 chars per param). Input validation is thorough with Zod schemas. However, error handling lacks recovery guidance, failures return generic messages without suggesting next steps. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite WRITE risk classification. Output schema is well-structured but could benefit from explicit error response documentation.

• Add tool annotations: mark 'render' with destructiveHint=true (stores files in R2) and idempotentHint=false (each call produces new render ID). Use @destructive/@idempotent JSDoc or MCP annotation syntax.
• Enhance error messages with recovery paths. E.g., 'Render timeout after 15s waiting for selector .content. Try: increase timeout, simplify selector, or verify HTML loads in browser.' instead of generic failure text.
• Document error response schema explicitly in outputSchema or add an 'errors' field to structuredContent showing error code, message, and suggested action.
• Add security note to tool description: 'Presigned URLs are valid for 24 hours and act as bearer tokens. Treat them as secrets, do not log or share. R2 storage is private to this server.'
• Clarify timing semantics: 'waitForSelector.timeout and waitForTimeout are independent. Total wait = min(selector.timeout, 15s) + waitForTimeout. If selector never matches, render fails after selector.timeout.'