RedNote MCP has critical definition gaps. Tool descriptions are present but minimal (10-20 chars in Chinese, below the 50-200 char LLM-optimized baseline). Input schemas are visible but lack depth: search_notes has 'keywords' (string) and optional 'limit' (number) with minimal descriptions; get_note_content and get_note_comments accept only 'url' (string); login has no visible input schema. No output schemas documented anywhere. Parameters lack type constraints (e.g., limit has no min/max bounds). Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. The login tool is particularly problematic: no description of what credentials it accepts, no security guidance on secret handling, no indication of side effects or state changes. Tool composition is weak, no clear chaining between search_notes (returns Note[]) and get_note_content/get_note_comments (both require URL), forcing agents to extract URLs manually. No pagination support despite search returning arrays. Overall, definitions are skeletal and would not pass code review.

• Add detailed English descriptions (100-150 chars) to all tools explaining WHAT they do, WHEN to use them, and WHAT they return. E.g., search_notes: 'Search RedNote (Xiaohongshu) for notes matching keywords. Returns up to [limit] results with title, author, URL, and engagement metrics. Use this to discover content before fetching details.'
• Document output schemas for all tools. search_notes should return {notes: [{title, author, url, likes, comments, ...}], total_count}. get_note_content should return {title, content, author, url, ...}. This enables LLM chaining.
• Add constraints to numeric parameters: limit should be 1-100 (default 10). Document in description: 'Maximum results to return (1-100, default 10).'
• Redesign login tool: move credentials to server-side environment variables (REDNOTE_USERNAME, REDNOTE_PASSWORD). Tool should accept no parameters and return {status: 'logged_in'|'failed', message}. Add description: 'Authenticate with RedNote using stored credentials. Must be called before other tools. Returns success/failure status.'
• Add error handling: return structured errors like {error: 'invalid_limit', message: 'limit must be 1-100, got 500', suggestion: 'Try limit=50'}. Document in tool descriptions: 'Returns error if keywords empty, limit invalid, or network fails. Errors are retryable unless marked fatal.'

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

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