Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output structure. All tools have descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization). Input schemas present but minimal, latitude/longitude have range constraints, but 'search' query lacks any validation guidance. No output schemas documented. Error handling returns text blobs without actionable recovery guidance. Tool composition is reasonable (get-alerts, get-forecast, search are distinct), but descriptions are too terse to guide LLM selection effectively.

• Expand tool descriptions to 50-200 chars. Example: 'Get active weather alerts for a US state (2-letter code: CA, NY, etc.). Returns event type, severity, area, and headline. Use this to check for warnings before get-forecast.'
• Document output schemas. Define that get-alerts returns {alerts: [{event, area, severity, status, headline}]}, get-forecast returns {location: {lat, lon}, periods: [{name, temp, wind, forecast}]}.
• Add error classification to responses: {error: 'not_found'|'invalid_input'|'service_unavailable', message: '...', recovery: 'Try search_users() first' or 'Retry in 30s'}.
• Add input validation: state parameter should validate against 50 US state codes (enum or regex); latitude/longitude already have range constraints, good. Query parameter should have max length (e.g., 500 chars).
• Cap get-forecast results at 10 periods and add pagination (offset, limit) to prevent context bloat. Document the limit in the description.