Two tools with identical schemas and reasonable descriptions. Both tools have complete input schemas with proper types, ranges, and defaults. Descriptions are adequate (120-150 chars) but lack actionable context about when to use each tool or what distinguishes them. Parameter descriptions are present but generic. No output schema documented. Error handling is minimal, only a generic 'Screenshot failed' message with no recovery guidance. The two tools appear to do nearly the same thing (take_screenshot vs screenshot_to_base64), which violates single-responsibility principle.

• Consolidate take_screenshot and screenshot_to_base64 into a single 'capture_website' tool with a 'return_format' enum parameter (image|base64). This eliminates ambiguity and reduces LLM decision overhead.
• Document output schemas explicitly: 'Returns ImageContent (mimeType: image/png|image/jpeg) for image formats, or TextContent with base64-encoded data and metadata (size, viewport, format) for base64 mode.'
• Enhance error messages with recovery guidance: 'Invalid URL format. Expected http:// or https://. Did you mean: https://example.com?' or 'Timeout waiting for selector .main-content (waited 30s). Page may not have loaded. Try increasing wait_for timeout or removing wait_for constraint.'
• Add parameter constraints to descriptions: 'wait_for: CSS selector (e.g., .content, #main). Timeout: 30s. If selector not found, screenshot proceeds anyway.' and 'inject_css: Custom CSS string (max 10KB). Injected into <head> before capture. Useful for hiding elements or adjusting layout.'
• Clarify use cases in tool descriptions: take_screenshot → 'Use when you need to display the screenshot directly to the user (returns image).' screenshot_to_base64 → 'Use when embedding the screenshot in a report, email, or passing to another tool that expects base64 data.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry