I gave my AI agent live aviation weather — building a free Aviation MCP server

#mcp #ai #aviation #javascript

I'm a commercial pilot who builds software. Last week I noticed something: ask any AI assistant "what's the weather at JFK right now and is it VFR?" and it either guesses, hallucinates a METAR, or tells you to go check a website. LLMs have no live aviation data.

So I built an MCP server that fixes that. It gives Claude, ChatGPT, Cursor — any MCP client — six aviation tools that return real data:

get_metar — current decoded METAR for any ICAO airport (flight category, wind, visibility, temp, dewpoint), optional TAF
get_airport — airport info by ICAO (name, IATA, city, coordinates, elevation, runways)
get_aircraft — aircraft specs by slug (engines, range, cruise, ceiling, MTOW, type rating)
get_glossary_term — definitions from an aviation glossary
practice_questions — FAA-style exam questions with answers
quiz_of_the_day — a daily aviation question

No API key. No signup. It's a thin MCP wrapper over a free aviation API I maintain (Rotate Pilot), so the tools are just typed HTTP calls — the hard part is the data, not the protocol.

Why MCP, and why this was easy

The Model Context Protocol is becoming the default way to hand tools to LLMs in 2026 — Claude, Cursor, Cline, Continue and Windsurf all speak it. If you have any API, wrapping it as an MCP server drops it into every AI client at once. That's a free distribution channel most API owners are sleeping on.

The server runs as an Apify Standby Actor (Streamable HTTP /mcp + legacy SSE), using the official @modelcontextprotocol/sdk. Each tool definition is ~10 lines: a name, a JSON schema, and a buildPath(args) that returns the API path. The server fetches it and returns the JSON. That's the whole thing.

Connect it

{
  "mcpServers": {
    "aviation": {
      "url": "https://renzomacar--aviation-mcp.apify.actor/mcp?token=YOUR_APIFY_TOKEN"
    }
  }
}

Then ask your agent: "Pull the METAR for KSFK and EGLL, tell me which is VFR, and compare a Cessna 172 to a Piper Warrior on cruise speed." It will call the tools and answer from real data.