Forecast Storage MCP Server

# Weather Agent Integration Plan ## Overview Replace filesystem-based caching with Cloud SQL MCP server for forecast storage. ## Current Architecture ``` Weather Agent ├── forecast_cache.py (filesystem scan) │ ├── get_forecast_from_cache() → scans OUTPUT_DIR │ ├── cache_forecast() → no-op (files already written) │ └── get_cache_stats() → scans directories │ ├── generate_audio.py │ └── Saves audio to local OUTPUT_DIR/{city}/ │ └── forecast_writer (implicit) └── Saves text to session state ``` ## New Architecture ``` Weather Agent ├── forecast_storage_client.py (NEW - MCP client wrapper) │ ├── upload_forecast_to_storage() → calls MCP upload_forecast │ ├── get_cached_forecast_from_storage() → calls MCP get_cached_forecast │ └── get_storage_stats_from_mcp() → calls MCP get_storage_stats │ ├── generate_audio.py (MODIFIED) │ └── Saves audio temporarily, returns path for upload │ └── agent.py (MODIFIED) └── Uses MCP client wrapper tools instead of forecast_cache ``` ## Changes Required ### 1. Create MCP Client Wrapper **File**: `weather_agent/forecast_storage_client.py` Wrapper functions that call MCP server tools: - `upload_forecast_to_storage(city, text, audio_path, timestamp)` - `get_cached_forecast_from_storage(city)` - `get_storage_stats_from_mcp()` ### 2. Modify generate_audio.py **File**: `weather_agent/sub_agents/forecast_speaker/tools/generate_audio.py` **Current behavior**: Saves directly to `OUTPUT_DIR/{city}/forecast_audio_{timestamp}.wav` **New behavior**: - Save to temporary location - Return path for agent to upload via MCP - Agent handles upload after both text and audio are ready ### 3. Update agent.py **File**: `weather_agent/agent.py` **Replace**: ```python from .forecast_cache import get_forecast_from_cache, cache_forecast, get_cache_stats ``` **With**: ```python from .forecast_storage_client import ( get_cached_forecast_from_storage, upload_forecast_to_storage, get_storage_stats_from_mcp ) ``` **Update instructions**: - Use `get_cached_forecast_from_storage` instead of `get_forecast_from_cache` - Use `upload_forecast_to_storage` after sub-agents complete - Handle MCP response format (JSON with status) ### 4. Configuration **File**: `.env` (project root) Add MCP server endpoint: ```bash # Option A: MCP Server runs as subprocess MCP_FORECAST_STORAGE_PATH=./forecast_storage_mcp/server.py # Option B: MCP Server runs externally MCP_FORECAST_STORAGE_URL=http://localhost:3000 ``` ## Migration Steps 1. ✅ Create MCP server (DONE) 2. ⏳ Create MCP client wrapper 3. ⏳ Modify generate_audio.py 4. ⏳ Update agent.py instructions and tools 5. ⏳ Test with sample forecast 6. ⏳ Deprecate forecast_cache.py (keep for reference) ## Backward Compatibility During transition: - Keep both systems (filesystem + Cloud SQL) - Add flag: `USE_CLOUD_SQL_STORAGE=true/false` - Gradual migration path ## Testing Strategy 1. **Unit tests**: Test MCP client wrapper functions 2. **Integration test**: - Generate forecast → Upload to Cloud SQL - Request same city → Retrieve from cache - Verify TTL expiration 3. **End-to-end test**: Full agent workflow with MCP storage ## Rollback Plan If issues arise: 1. Set `USE_CLOUD_SQL_STORAGE=false` 2. Agent falls back to filesystem cache 3. Debug MCP integration separately