Both tools have descriptions in Korean, but lack English descriptions and proper parameter documentation. analyze_us_stock has a single string parameter with minimal description; get_global_top_news has no parameters but no output schema documented. Neither tool declares return types or error recovery guidance. Descriptions are present but non-English and lack LLM-optimization for English-speaking agents. No tool annotations (readOnlyHint, etc.) despite both being read-only. Tool names follow verb_noun pattern (analyze_, get_) but descriptions do not explain WHEN to use each tool or what distinguishes them.

• Add English descriptions (50-200 chars) for both tools explaining WHAT they do, WHEN to use them, and what they return. Example: 'analyze_us_stock: Fetch real-time stock price and AI-analyzed news sentiment for a US ticker. Returns formatted report with price, change%, and news summaries with sentiment/impact scores. Use to get current market data and news analysis for a specific stock.'
• Document output schema for both tools. Define return type as object with fields: title (string), price (number), currency (string), change_percent (number), news_items (array of {title, sentiment, impact, summary}). Include example output.
• Add format constraint to ticker parameter: 'US stock ticker symbol (1-5 uppercase letters, e.g. AAPL, NVDA, TSLA). Invalid tickers return error with suggestion to search available symbols.'
• Implement error categorization: return structured error objects with {error_type: 'not_found'|'invalid_input'|'service_error', message: '...', recovery_hint: '...'}. Example: 'Ticker INVALID not found. Try search_tickers(query) to find valid symbols.'
• Add readOnlyHint annotation to both tools in FastMCP decorator: @mcp.tool(annotations={'readOnlyHint': True}) to signal safety and enable agent retry logic.