Server exposes 6 tools with complete JSON schemas and descriptions. Tool names follow verb_noun convention (list_, get_, create_, update_, delete_, describe_). Descriptions are present and adequate (50-150 chars). However, parameter descriptions are minimal or generic ('Title field', 'Body field', 'Status field'), lacking context about constraints, formats, or when to use each field. The describe_resources tool is well-designed for agent self-discovery. Error handling is present but not documented in tool descriptions. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Output schemas are not explicitly documented in tool definitions.

• Add explicit output schema documentation to each tool. For list_articles, document: {items: [{id, title, body, status, created_at}], total: int, page: int, per_page: int}. For create_article, document: {id, title, body, status, created_at}.
• Enhance parameter descriptions with constraints and examples. E.g., 'status' → 'Status field (enum: draft, published, archived)'. 'per_page' → 'Records per page (1-100, default 25)'.
• Add tool annotations to all tools. Mark list_articles and get_article with readOnlyHint=true. Mark create_article, update_article with no hint. Mark delete_article with destructiveHint=true.
• For delete_article, add a confirmation step or dry-run parameter. E.g., 'confirm: boolean (default false), if false, returns what would be deleted without executing'.
• Document error scenarios in tool descriptions. E.g., 'Returns 404 if article not found. Try list_articles to verify the ID exists.' This guides agent recovery.