Server has 4 well-named tools with clear verb-noun patterns (export_article, get_article_info, list_exports, check_translation). Descriptions are present and reasonably detailed (100-200 chars), meeting baseline expectations. Input schemas are visible with type definitions and parameter descriptions. However, output schemas are not formally documented, responses are text-based summaries rather than structured JSON objects. Error handling returns text messages but lacks actionable recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Parameter descriptions could be more precise about constraints and formats.

• Convert text-based responses to structured JSON objects. export_article should return {file_path, title, author, page_count, image_count, word_count, translated_to?, validation_ok, warnings?, errors?}. get_article_info should return {title, author, published_at, modified_at, block_count, image_count, word_count, cover_image_url?, block_types}. This enables reliable field extraction and downstream chaining.
• Add tool annotations via mark3labs/mcp-go: export_article.destructiveHint=true (writes to disk), others with readOnlyHint=true. This signals operation safety to agents without requiring description parsing.
• Enhance error messages with actionable recovery. E.g., 'Invalid article URL: expected https://x.com/user/article/123 format. Check the URL and try again.' or 'Failed to fetch article: network timeout. Retry in 30 seconds or check your internet connection.'
• Add enum constraint to 'translate' parameter: restrict to ISO 639-1 codes (de, fr, ja, es, etc.) or document the exact set supported by the Ollama model. Validate server-side and return 'Unsupported language: got "english", must be ISO 639-1 code (e.g., en, de, fr).'
• Add min/max bounds to 'limit' parameter in list_exports: 1 - 1000 (or appropriate max). Document in description: 'Maximum number of files to return (1 - 1000, default 20).' Validate and reject out-of-range values with clear error.