Single tool 'historical_candles' has a valid input schema with three string parameters (symbol, from_date, to_date), each with descriptions. However, the tool description is in Chinese and mixes implementation details ('優先使用本地數據，如果本地沒有再使用 API') with parameter documentation. No output schema is documented, the function returns a dict with 'status', 'data', 'message' fields, but the structure of 'data' (list of dicts with OHLCV fields) is never formally specified. Parameter descriptions lack format constraints (e.g., date regex, symbol format rules beyond examples). No error handling guidance, failures return generic error dicts without recovery hints. Tool name 'historical_candles' is acceptable (verb_noun pattern) but lacks clarity on whether it returns candles, bars, or raw OHLCV. The resource 'twstock://{symbol}/historical' is defined but never integrated with the tool.

• Document the output schema explicitly: 'Returns {status: string, data: [{date: string (ISO 8601), open: float, high: float, low: float, close: float, volume: int, vol_value: float, price_change: float, change_ratio: float}], message: string}'. Include field types and meanings.
• Rewrite tool description in English: 'Fetch historical OHLCV candles for a Taiwan stock symbol within a date range. Prioritizes local cached data; falls back to API if cache is incomplete. Returns up to 365 days per request; longer ranges are auto-segmented.'
• Add format constraints to parameters: symbol as enum or regex (e.g., '^[0-9]{4,5}$'), from_date/to_date as ISO 8601 strings with min/max bounds (e.g., 'from_date must be >= 1990-01-01 and <= today').
• Implement error categorization: return {status: 'error', code: 'symbol_not_found|date_range_invalid|api_unavailable', message: '...', recovery: 'Try search_symbols() first' or 'Reduce date range to <= 365 days'}.
• Add a discovery tool (e.g., 'list_symbols' or 'search_symbols') to help agents find valid stock codes before calling historical_candles. This prevents wasted calls with invalid symbols.