Why We Built an MCP Server for Drones

The first drone platform built for AI agents.

Every commercial drone flight starts the same way. You open B4UFLY to check if you're in controlled airspace. Then you pull up Google Earth to look at the site. Then you check the FAA's TFR map to make sure nothing popped up overnight. Then you open your flight planner to build waypoints. Then you check the weather. Then you export to your controller's format.

Six tools. Six tabs. Every single job.

I've been doing this for years, and it never stopped being tedious. The information exists -- it's all public FAA data -- but it's scattered across a half-dozen apps and websites that don't talk to each other. Pre-flight planning takes longer than most flights.

So I built something different.

Another app is not the answer

The drone industry's solution to every problem is another app. Another subscription. Another login. Another interface to learn.

I didn't want to build another app. Dronelytics already has a free airspace map that shows every layer a Part 107 operator needs -- UASFM ceilings, class airspace, TFRs, NOTAMs, SUAs, airports, Mode C veils, military training routes, the works. But even a good app is still an app. You still have to open it, navigate to your location, and eyeball the layers yourself.

What I actually wanted was to ask a question and get an answer.

"Can I fly at 123 Main Street tomorrow at 2pm?"

Not "here's a map, go figure it out." An actual answer: yes, no, or here's what you need to know first.

That's not a UI problem. That's an AI problem.

AI agents need real-world tools

You've probably used Claude or ChatGPT to help with proposals, scheduling, or client emails. These AI assistants are genuinely useful for text work. But ask one "can I fly my drone at the Golden Gate Bridge?" and you'll get a confident-sounding answer that's probably wrong. They don't have access to live FAA data. They're guessing based on training data that might be months or years old.

AI agents become actually useful when they can call real tools with real data. Not hallucinate an answer -- query a live database and return facts.

That's what MCP does.

MCP stands for Model Context Protocol. It's an open standard that lets AI assistants connect to external tools. Think of it like USB for AI -- a universal way for any AI agent to plug into any data source or service.

When an AI assistant has access to an MCP server, it can call specific tools on your behalf. Instead of guessing about airspace, it queries the actual FAA data. Instead of making up a flight plan, it runs the real waypoint generation math.

What the Dronelytics MCP server does

I built 24 tools that cover the full pre-flight workflow. Here's what matters:

Airspace intelligence

The check_airspace tool queries 11 FAA data layers at any lat/lng in the US and returns a structured flyability assessment. Airspace class, UASFM ceiling, active TFRs, NOTAMs, special use airspace, NSUFR restrictions, nearby airports, Mode C veil status, stadiums, FRIAs, and military training routes. One query, complete picture.

You can also check live ADS-B traffic, pull active TFRs, or get NOTAMs near any location.

Flight validation

Give it a location, altitude, and time, and validate_flight checks everything against Part 107 rules. Altitude limits, UASFM ceilings, active TFRs with time overlap, NSUFR restrictions, LAANC requirements, daylight rules. Returns PASS, CAUTION, or FAIL with specific rule citations.

Pre-flight briefing

This is the one I'm most proud of. Point preflight_briefing at any saved mission and it combines the mission data with live airspace intelligence to produce a GO, CAUTION, or NO-GO recommendation. It checks everything -- airspace class, ceiling violations, TFRs, NOTAMs, nearby airports, Mode C veil, stadiums, even live aircraft traffic.

Mission generation

Five automated survey types: grid mapping, orbit/POI, corridor, spiral, and panorama. The grid planner calculates optimal lawnmower patterns based on your camera, altitude, and overlap settings. Tell your AI agent "plan a grid survey of this 10-acre field at 200 feet with my Mavic 3" and it generates the waypoints.

Mission management

Full CRUD -- create, read, update, delete, duplicate, share. Import Litchi CSV files. Export to KML, GPX, QGroundControl, Litchi CSV, or DJI WPML. Your AI agent can manage your entire mission library.

What this looks like in practice

Here's a real example. You're sitting in your truck before a roof inspection job. You open Claude and type:

"Can I fly at 4521 NE Sandy Blvd, Portland?"

Claude calls check_airspace, checks all 11 layers, and responds:

"Yes, you can fly there. It's Class G airspace with a 400ft ceiling. No active TFRs. Portland International is 4.2nm northeast -- maintain awareness. You're inside the Portland Mode C veil, so ADS-B Out is required. Two NOTAMs active in the area but neither is UAS-related."

Then you say: "Plan a grid survey at 200 feet with 80% front overlap using my Mavic 3, and give me a pre-flight briefing."

Claude generates the waypoints, creates the mission, runs the briefing, and gives you a GO with notes about the nearby airport.

"Export that to Litchi."

Done. CSV ready. The entire pre-flight workflow in three messages.

How to set it up

Two minutes. You need a Dronelytics account (free) and an API key.

1. Go to hub.dronelytics.io, sign up, and go to Settings > API Keys to generate a key.
2. Add this to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "dronelytics": {
      "command": "npx",
      "args": ["-y", "@dronelytics/mcp"],
      "env": {
        "DRONELYTICS_API_KEY": "dk_your_key_here"
      }
    }
  }
}

3. Restart Claude Desktop. That's it.

Works with Claude Code, Cursor, Windsurf, and any other MCP-compatible client. Same config.

Why it's free

I'm a solo founder building Dronelytics in public. This isn't a VC-backed company with a monetization deadline. The MCP server is free because right now I care more about people using it than paying for it.

Every time an AI agent checks airspace through Dronelytics, that's a pilot who just discovered the platform. Every query is a brand impression that happens inside someone's workflow, not in an ad they're scrolling past. That's worth more to me than $29 a month from a handful of early adopters.

There will be paid tiers eventually -- mission generation and export are expensive to run at scale. But the core "can I fly here?" question should always be free. That's the whole point.

What's next

This is version 0.3.0. Here's what I'm working on:

If you're a developer building AI agents and want to add drone airspace awareness as a capability, the npm package is @dronelytics/mcp. If you're a drone operator who uses Claude or ChatGPT, try asking it to check airspace for your next job.

I'd genuinely love to hear what you think. Find me on Reddit or open an issue on GitHub.

Mark builds Dronelytics, a mission planning platform for commercial drone operators. He holds a Part 107 certificate and has strong opinions about pre-flight checklists.