AdSense MCP has well-structured tool definitions with consistent naming (all start with get_/list_), clear descriptions (avg 120 chars), and complete input schemas with enums and type constraints. However, output schemas are entirely undocumented, LLMs cannot predict response structure or plan downstream calls. Parameter descriptions are minimal (avg 40 chars) and lack format guidance. No error handling guidance, no pagination documentation despite tools returning lists, and no security/permission declarations. Tool composition is sound (single responsibility), but missing critical production patterns like result limits, chaining IDs, and recovery hints.

• Document output schema for every tool. Include field names, types, and sample structure. E.g., get_earnings_summary returns {earnings: number, page_views: number, impressions: number, clicks: number, ctr: number, rpm: number}.
• Add pagination support to list-returning tools: accept limit (1-100, default 20) and offset/cursor params; return total_count and next_cursor in response.
• Expand parameter descriptions to 60-100 chars. Include format examples, valid ranges, and what happens on invalid input. E.g., 'Start date in YYYY-MM-DD format (e.g., 2024-01-15). Must be ≤ end_date. Returns error if format is invalid.'
• Add error handling guidance to each tool description. E.g., 'Returns 401 if AdSense credentials are invalid. Returns 400 if date_range is custom but start_date/end_date are missing. Returns 429 if API quota exceeded, retry after 60 seconds.'
• Declare required permissions for each tool. E.g., get_earnings_summary requires 'read:earnings'; get_payments requires 'read:payments'. Add a permissions field to tool metadata.