Server has 8 tools with consistent naming (verb_noun pattern: velog_*), detailed descriptions in Korean, and explicit input schemas with types and descriptions. Tool annotations (READ_ONLY, WRITE, DESTRUCTIVE) are properly declared. However, output schemas are not documented in the code, responses are shaped ad-hoc via _shape_post() helper without formal schema declarations. Error handling is present but generic (ToolError wrapping). Parameter descriptions are strong (e.g., timeout_sec with min/max), but some tools lack comprehensive validation guidance. No pagination metadata (total_count, next_cursor) visible in list tools despite accepting cursor parameter.

• Add formal JSON Schema output declarations for all tools. Document response fields (post_id, title, url, tags, body, etc.) with types and descriptions so LLMs can plan chaining.
• Enhance pagination: document next_cursor and total_count in velog_list_posts and velog_list_series responses. Add example: 'Returns up to 50 posts; use cursor from last item to fetch next page.'
• Wrap VelogError exceptions with actionable recovery hints. E.g., 'Auth failed: token expired. Call velog_login() to refresh.' instead of generic ToolError.
• Add explicit validation for file_path in velog_publish_markdown_file: check Path.is_absolute() and Path.exists() before processing. Return clear error if path is relative or missing.
• Document dry_run mode behavior in velog_whoami and velog_publish_post descriptions. Clarify that dry_run=true prevents actual API calls and returns mock responses.

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

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