Server has 5 tools with explicit schemas and descriptions. Naming follows verb_noun convention (velog_whoami, velog_list_posts, velog_get_post, velog_write_post, velog_edit_post). Descriptions are present and actionable (avg ~100 chars). Schemas use Zod with type constraints. However, parameter descriptions are sparse or missing for several fields (e.g., 'tag' in velog_list_posts has no description; 'meta' object lacks field documentation). Output schemas are not documented, callers must infer structure from GraphQL queries. Error handling is minimal: no recovery guidance, no categorization of retryable vs fatal errors. No tool annotations (readOnlyHint/destructiveHint). Write tools (velog_write_post, velog_edit_post) lack confirmation/dry-run patterns despite modifying state.

• Document output schemas for all tools. Example: velog_list_posts returns {posts: [{id, title, url_slug, short_description, thumbnail, released_at, updated_at, is_private, is_temp, likes, comments_count, tags, user: {username, profile: {display_name}}}]}. Include field types and descriptions.
• Add descriptions to all parameters. 'tag' should be 'Filter posts by tag name (e.g., "javascript", "react"). Case-insensitive.' 'meta' should document expected object structure (e.g., 'Custom metadata object; structure depends on Velog version').
• Add tool annotations: mark velog_write_post and velog_edit_post with destructiveHint=true; mark read tools with readOnlyHint=true. This enables clients to apply appropriate safeguards.
• Implement dry-run for write tools. Add optional 'dry_run' parameter to velog_write_post and velog_edit_post. When true, validate input and return a preview without publishing. Agents can review before committing.
• Enhance error messages with recovery guidance. Example: 'VELOG_ACCESS_TOKEN not set. Set the token in the MCP server environment or call velog_whoami to verify authentication.' Categorize errors as retryable (network timeout) vs fatal (invalid token).