Single tool 'get_weather' has a basic description but lacks critical specification details. Input schema is present with minimal parameter documentation. No output schema is documented. The tool description is only 59 characters, below the 194-char baseline for production tools. Parameter 'city' has a description but lacks constraints, format guidance, or error handling context. No error recovery guidance provided. The code shows no input validation, no enum constraints, no pagination support, and secrets (WEATHER_API_KEY) are hardcoded in source rather than injected. Tool does not follow the pattern:tool-chain or mxe:natural-identifiers patterns, no fallback suggestions if city is not found.

• Expand tool description to 100-200 chars: 'Fetch current weather for a specified city using WeatherAPI.com. Returns temperature, humidity, wind speed, and condition. Use this when you need real-time weather data. Provide the city name (e.g., Paris, New York, Tokyo).'
• Document full output schema: {city: string, region: string, country: string, local_time: string (ISO 8601), latitude: number, longitude: number, temperature_c: number, feelslike_c: number, humidity: number (0-100), wind_kph: number, wind_dir: string, pressure_mb: number, uv_index: number, visibility_km: number, condition: string, condition_icon: string (URL)}
• Add constraints to 'city' parameter: 'City name (1-100 characters, alphabetic and spaces only). Examples: Paris, New York, Tokyo, São Paulo. If the exact city is not found, the tool will return an error suggesting similar city names.'
• Move WEATHER_API_KEY to environment variable: `WEATHER_API_KEY = os.getenv('WEATHER_API_KEY')` with explicit error if not set.
• Implement differentiated error responses: 'City not found. Available matches: [list]. Try again with exact name.' vs 'API rate limit exceeded. Retry in 60 seconds.' vs 'Network timeout. Retry this request.' so LLMs know which errors are retryable.
• Add input validation: validate city name length (1-100), reject non-alphabetic characters except spaces/hyphens, check for SQL injection patterns, return actionable error messages.
• Remove or document the SSE endpoint. If streaming is needed, add parameters for update_interval_seconds and max_updates to prevent infinite streams. Document when to use streaming vs single fetch.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry