velog-mcp has 15 tools with minimal schema visibility and inconsistent documentation. Tool names follow verb_noun convention (get_, create_, search_) which is good. However, critical issues: (1) No input schemas visible in provided source code, cannot verify parameter types, constraints, or descriptions. (2) Tool descriptions are extremely brief (10-30 chars), far below the 50-200 char baseline for LLM optimization. E.g., 'velog_get_posts' → '벨로그 글 목록 조회' (11 chars), 'velog_search_posts' → '벨로그 글 검색' (8 chars). (3) No output schema documentation visible. (4) No error handling guidance in descriptions. (5) WRITE tools (velog_create_draft, velog_publish_post, velog_edit_profile, velog_upload_image) lack confirmation/dry-run patterns. The server has thoughtful security design (VELOG_ALLOW_PUBLIC flag, rate limiting, permission gates in code) but this is not reflected in tool definitions themselves.

• Expand all tool descriptions to 50-200 characters. Example: 'velog_get_posts' → 'Retrieve a paginated list of blog posts. Supports filtering by tag, author, and date range. Returns post metadata including title, excerpt, and view count. No authentication required.'
• Document input schemas for all tools with explicit parameter types, constraints, and descriptions. E.g., velog_search_posts should declare: query (string, required, 1-100 chars), limit (integer, 1-50, default 20), offset (integer, ≥0, default 0).
• Document output schemas showing field names, types, and which fields are needed for tool chaining. E.g., velog_get_posts returns [{ post_id, title, excerpt, author_id, created_at, ... }] so agents can pass post_id to velog_publish_post.
• Add confirmation/dry-run pattern to WRITE tools. E.g., velog_publish_post should accept dry_run=true to preview changes before is_private:false commit. Irreversible operations need agent review.
• Add error handling guidance to descriptions. E.g., velog_search_posts: 'Returns empty array if no matches. If authentication fails, error message will suggest VELOG_REFRESH_TOKEN setup. Rate limit errors are retryable after 60s.'