Scoring was not performed

• Document output schema for each tool. E.g., 'Returns {series_id: string, data: {id: string, title: string, units: string, frequency: string, data: [{date: string, value: string}]}}'. This enables chaining and extraction.
• Expand parameter descriptions to 50-150 chars with context. E.g., 'observation_start: Start date for observations to retrieve (format: YYYY-MM-DD). Only observations on or after this date are returned. Use with observation_end to filter time ranges.'
• Add recovery guidance to error responses. E.g., catch BaseFredAPIError and return {'error': 'Series ID not found. Valid series IDs are 3-letter+ codes (e.g., GDP, UNEMPLOYMENT). Use FREDSeriesSearch() to find available series.'}
• Distinguish similar tools in descriptions. FREDSeriesTags: 'Get metadata tags assigned to a specific series (e.g., seasonal adjustment, frequency). Use to explore what attributes a series has.' vs FREDSeriesSearchTags: 'Get tags that describe series matching a search query. Use to narrow search results by tags.'
• Cap FREDSeriesObservations and FREDSeriesSearch max results at 100 by default, document in description, and explain pagination. E.g., 'Returns up to 100 results per call (limit: 1-100). Use offset to page through results. Always check count field to see if more results exist.'
• Add state to descriptions. E.g., FREDSeriesSearch: 'Search FRED database for series by title, description, or series ID. Returns high-level metadata for each match. Call FREDSeriesObservations to fetch actual data points.'