Velog-mcp has well-structured tool definitions with consistent naming (verb_noun pattern) and complete Zod schemas for all 10 tools. Descriptions are present and actionable (avg ~80 chars). However, output schemas are not documented, responses are JSON.stringify'd without field documentation, forcing LLMs to infer structure. Error handling is minimal: auth failures return plain text, API errors are not categorized or recovery-guided. Parameter descriptions are adequate but lack format constraints (e.g., no regex for url_slug, no guidance on post ID format). Destructive operations (delete_post) lack confirmation patterns. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels.

• Document output schemas for all tools. For get_user_posts, specify: { posts: Array<{ id, title, url_slug, author, createdAt, likes, comments }>, nextCursor?: string }. For read_post: { id, title, body, author, createdAt, comments: Array<{ id, author, text }>, metadata }.
• Add tool annotations to server.tool() calls. Use readOnlyHint: true for get_* and search_* tools; destructiveHint: true for delete_post; idempotentHint: true for edit_post (if idempotent by design).
• Implement confirmation pattern for delete_post. Add a dry_run parameter (default false) or split into delete_post_confirm + delete_post_execute. Return { confirmed: true, postId, title } before actual deletion.
• Enhance error handling with categorization. Catch API errors and return: { isError: true, errorType: 'auth_required|not_found|rate_limited|server_error', message: '...', recoveryHint: 'Call login() first' or 'Retry in 60s' }.
• Add format constraints to parameter descriptions. For url_slug: 'URL-safe slug (lowercase, hyphens, 3-100 chars, no spaces)'. For post ID: 'UUID format (36 chars)'. For series_id: 'UUID or numeric ID from get_series_list()'.