DFlow MCP Server has 15 tools with inconsistent quality. Naming is generally strong (verb-noun pattern: get_event, get_markets, etc.), but schemas are only partially visible in the provided source. Descriptions are present and moderately detailed (50-150 chars typical), but several critical issues emerge: (1) output schemas are not documented anywhere in the source provided, forcing the score down significantly; (2) some tools lack proper constraint documentation (e.g., percentiles parameter in get_forecast_percentile_history is described as 'Comma-separated list' but no enum or pattern is enforced); (3) error handling guidance is absent, no mention of recovery steps or error categorization; (4) parameter relationships are undocumented (e.g., get_events has both 'seriesTickers' and 'status' filters but no guidance on how they interact). Tools 1-15 show consistent patterns but lack depth in documentation that production-grade tools provide. The web-server.ts shows only 3 tools defined (get_events, get_markets, get_trades) while the spec lists 15, suggesting tool definitions may be spread across multiple files or not fully visible in the sample code provided.

• Add documented output schemas to all 15 tools. Define the fields returned by each API call (e.g., get_event returns {event_id, series_ticker, subtitle, markets: [{market_id, strike, ...}], ...}). Use a structured comment or inline JSDoc schema.
• Add error handling guidance to tool descriptions. Example: 'Returns event metadata. If event_id not found, try get_events() to discover valid IDs. On timeout, retry with exponential backoff.'
• Replace free-form 'Comma-separated list' parameter descriptions with formal constraints. For percentiles, add enum: ['1', '5', '10', '25', '50', '75', '90', '95', '99'] or a regex pattern /^\d+(,\d+)*$/ with description 'Comma-separated integers between 1-99'.
• Document parameter relationships explicitly. In get_events description, add: 'Filters combine with AND logic: seriesTickers AND status AND isInitialized all must match. Omit filters to retrieve all events.'
• Standardize pagination cursors across tools. Choose one type (integer offset or string token) and use it consistently. Document max limit (e.g., limit ≤ 100) and add total_count to all paginated responses.
• Add LLM-friendly discovery hints. Update get_live_data and get_live_data_by_event descriptions with: 'Use get_live_data for specific milestone IDs. Use get_live_data_by_event to fetch all milestones for an event and discover milestone IDs.'
• Increase time-series parameter descriptions. In get_event_candlesticks and similar, clarify what 'periodInterval' means: 'Candlestick interval in seconds (e.g., 60 for 1-minute, 3600 for 1-hour). Must align with data granularity available from the API.'