Aviation weather is understandably forward looking; the weather ahead of us is far more important than the weather behind us. Even so, many of us are pattern-matchers at heart and are curious:
What was the weather like this time last year, right here?
Is this a typical storm pattern for this area of the United States?
Can I expect the high pressure to wash out the crud anytime today?
Is it always this windy in southern Wyoming? (yes, it is)
So many years ago, long before everyone fell in love with AI, I built a weather tool to answer those questions which I could not find answers to anywhere else, and http://www.AWOSLOG.com was born.
Then, on a whim, I taught AI (an LLM) how to use it.
This is not a post about artificial intelligence replacing pilots, dispatchers, or briefers — it is a story about a scratch builder scratching an itch, one piece at a time, until the pieces started fitting together in ways that surprised even me.
The Itch
If you fly in Colorado or anywhere in the mountain west, you already know the drill. Check the METARs, check the TAFs, check the winds aloft, pull up the TFRs, then cross-reference the ceiling and visibility against the terrain you intend to fly through. Do this for your departure field, your destination, and every alternate you might need, and if you are lucky the weather might actually be similar to the briefing.
It is not difficult work, but it can be tedious work, and tedious work invites complacency. You check one field and assume the next one is similar, glance at the ceiling and skip the trend, or forget to check the TFRs entirely because you checked them yesterday and there were none.
What I wanted was a single place to look at current & historical weather across many stations simultaneously, with enough history to see the trends — not just a snapshot, but a story of how conditions evolved throughout the day. So I built AWOSLOG.com.
What AWOSLOG Does
AWOSLOG continuously collects METAR observations from over 3,000 airports across the United States, stores them in a database with up to a full year of history, and presents them through a web interface where you can pull up any station, see current conditions, and scroll through the history to watch trends unfold. It also shows live aircraft positions, active TFRs, and flight category status (VFR, MVFR, IFR, LIFR) at a glance.
None of this is revolutionary — I built this tool for me, and it is just a clean, fast, purpose-built tool that gives me the data I needed without the clutter. But, if the nicely charted summer density altitudes in Telluride or Leadville don't get your attention, nothing will.
Enter the Robot
Earlier this year, a protocol called MCP (Model Context Protocol) emerged as a standard way for AI assistants to connect to external data sources. Think of it as a universal adapter — a way for an AI like Claude to reach out and query live systems in real time, rather than answering from stale training data alone. This caught my attention, because an AI assistant that can look up the current weather at KASE is fundamentally more useful than one that can only recite what it learned about weather patterns months ago. The difference is the same as asking a pilot who flew the route this morning versus one who flew it last year; both points of view are useful but are very different.
I added an MCP server to AWOSLOG, and while the implementation details are not important here, what it enables is worth talking about.
What It Looks Like
Once connected — a one-time setup that takes about two minutes — you can ask Claude questions the same way you would ask a briefer:
"What's the weather at Montrose? How does today differ from last years average?"
Claude queries awoslog.com in real time and returns the current METAR in plain English, including wind, temperature, visibility, ceiling, and pressure. But it goes well beyond a simple lookup.
"Compare the weather at KASE and KDEN today and last week."
Claude pulls both stations and gives you a side-by-side comparison, calling out the differences that matter for your decision-making.
"What was the wind trend at Gunnison over the last 24 hours?"
Claude retrieves the full observation history and summarizes the trend, noting when the wind picked up, when it shifted, and when it calmed down.
"I'm flying a C180 into Telluride this afternoon. What should I be concerned about?"
This is where it gets genuinely useful. Claude checks the current conditions, evaluates the wind against typical crosswind limitations for a Cessna 180, looks at the ceiling relative to the terrain, checks for TFRs along your likely route, and assembles a coherent assessment in a few seconds. Not a go/no-go decision — that is always yours — but a solid starting point built from live data.
"Are there any aircraft near Aspen right now?"
Claude checks the live ADS-B feed and tells you what's in the area, including callsigns, altitudes, and aircraft types.
Why This Matters
I am not suggesting that any pilot should replace their weather briefing process with an AI conversation — standard briefing resources, 1800wxbrief, and good old-fashioned judgment are not going anywhere. But the barrier between a pilot and their data is unnecessarily high, and it has been that way for a long time. We have the data, and we have the systems that collect and store it. What we have lacked is a natural way to ask questions about it.
When you call Flight Service, you are talking to a human who has access to the data and the experience to interpret it. MCP gives an AI assistant that same access, and while it does not replace the experience, it does provide instant availability at any hour from anywhere. That is valuable for backcountry pilots checking conditions at a remote strip before a long drive to the airfield, for instructors preparing a lesson on weather decision-making, and for anyone who has ever thought "I just want to know if I can fly today" and then spent fifteen minutes clicking through four different websites to piece together an answer.
Try It
The MCP connection works with Claude Desktop (the installed application for Mac or Windows) or Claude Code (the command-line tool), though it does not work with the free web version of Claude at claude.ai.
Setup takes two minutes, and the full instructions including configuration details and example questions are available in the PDF linked below:
The short version is straightforward: install Claude Desktop from [claude.ai/download](https://claude.ai/download), open Settings, go to Developer, click Edit Config, and add this to the configuration file:
json
{
"mcpServers": {
"awoslog": {
"type": "streamable-http",
"url": "http://awoslog.com/mcp"
}
}
}
Restart Claude Desktop and start asking questions about weather. The data is live, the source is the same NOAA METAR feed used by every other aviation weather tool, and the AI is not making up weather — it is reading the same observations you would read yourself and helping you make sense of them.
What's Next
AWOSLOG is an ongoing project, and the MCP integration is the latest addition. There is more in the works, including community-operated weather stations designed for airstrips that currently have no reporting at all — small solar-powered units that upload observations directly to awoslog.com so pilots finally have wind and weather data at fields that have historically offered nothing but a windsock and a guess.
But, all of that is a story for another day...