Two tools with basic verb-noun naming (get_alerts, get_forecast) and minimal descriptions. Both tools have input schemas with types and descriptions, but output schemas are undocumented (return plain strings). Descriptions are 50-60 chars, meeting minimum length but lacking actionable context for LLM selection. No error handling guidance, no parameter constraints (e.g., latitude/longitude bounds), no documentation of what the string output contains. Tools are READ_ONLY but lack explicit idempotent hints. No pagination, no structured output, no recovery guidance.

• Add structured output schemas. get_alerts should return {alerts: [{event, severity, area, description, instructions}], count: int}. get_forecast should return {periods: [{name, temperature, unit, wind, forecast}], location: {lat, lon}}.
• Expand descriptions to 100-150 chars with context: 'Get active weather alerts for a US state (severe storms, floods, etc.). Returns empty if none active. Use before get_forecast to check for urgent warnings.'
• Add parameter constraints: state as enum of valid US state codes, latitude/longitude with min/max bounds, and format hints in descriptions.
• Implement error handling with recovery guidance. Catch API timeouts and return: 'NWS API timeout. Retry in 30 seconds or use cached forecast from last 6 hours.'
• Add tool annotations (readOnlyHint: true, idempotentHint: true) to both tools so agents know they are safe to retry.