AI-Alert
vmihalache/AI-Alert
Local AI agent for US weather alert monitoring using MCP, PostgreSQL and NWS API
Instalação
npx convez-ai vmihalache/AI-Alert1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
AI-Alert
Company: AI-Alert
Scope: Provides state-level weather alert monitoring for the US based on user queries.
Live API: https://ai-alert-production.up.railway.app/api/weather
Overview
A cloud-deployed AI agent system that draws results from the NWS (US National Weather Service) API to show relevant advice and alerts to people who want to know if it is safe to travel through a certain US state.
The system uses a Model Context Protocol (MCP) server to connect a cloud-hosted AI model (Groq / Qwen3-32B) to a PostgreSQL database, enabling natural language queries about active weather alerts across the United States.
Domain Rules
- Results are shown only if the NWS API returns data for the requested state
- Queries are reliable only if they mention the state name — not lower level locations or latitude/longitude geo information
- Queries are not reliable if the user wants reports based on complex dates or timelines
- Response quality depends on the AI model and may vary between runs
Use Cases
UC1 — User receives responses based on queries that include a State Name ✅ Implemented
Our product takes the user's input, extracts the state name from it and performs a query to the NWS API. It then shows the results, if any.
UC2 — The Agent recognizes different location input types besides State Name 🔲 Pending
Covers cities, villages and latitude/longitude geo information provided by the user.
UC3 — User can request a report for the last month to see the weather evolution 🔲 Pending
Allows the user to query historical weather data and observe alert trends over time.
Test Cases
UC1 — User receives responses based on queries that include a State Name
- TC-001 — Should return a response for a state with active alerts
- TC-002 — Should return a response for a state with no active alerts
- TC-003 — Should handle Washington DC as a valid location
- TC-004 — Should store ingested alerts in the database with the correct state code
- TC-005 — Should not store duplicate alerts on repeated ingestion
- TC-006 — Should return alerts filtered by the correct state
UC2 — Agent recognizes different location input types 🔲 Pending
- TC-007 — Should handle a city name as input
- TC-008 — Should handle latitude and longitude as input
UC3 — User can request a report for the last month 🔲 Pending
- TC-009 — Should return historical alerts for a given state
- TC-010 — Should show weather evolution over the requested time period
Implementation Status
| Component | Status |
|---|---|
| MCP Pipeline with Local AI | ✅ Implemented |
| MCP Pipeline with Cloud AI (Groq) | ✅ Implemented |
| Express API Server | ✅ Implemented |
| Railway Deployment | ✅ Live |
| Neon PostgreSQL | ✅ Live |
| React Frontend | 🔲 Not done |
Usage
Send a POST request to the live API:
POST https://ai-alert-production.up.railway.app/api/weather
Content-Type: application/json
{
"question": "What is the weather in Virginia?"
}
Stack
TypeScript · Playwright · PostgreSQL (Neon) · Drizzle ORM · MCP SDK · Groq (Qwen3-32B) · NWS API · Express · Railway