DoorSense

Ask Alexa "Is Mom okay?" and get a real answer from her front door. DoorSense turns doorbell events into a wellness check for parents living alone.

Comment

Inspiration

A lot of us have a parent or grandparent who lives alone in another city. You call in the evening, nobody picks up, and for the next twenty minutes you're running through worst cases in your head. The thing is, the front door already knows most of what you want to know. Did she go out today? Did the neighbor stop by? Has the same guy rung the bell three times? Is the package from this morning still sitting outside? None of that is hard data to get. It just isn't in a form you can ask about. So I built DoorSense. You ask Alexa a normal question about your mom, and it answers from what happened at her door.

What it does

You can talk to it — "did anyone visit Mom today?", "should I be worried about that guy?" — and it answers in a couple of sentences, out loud.

Behind that is a simple wellness score. It drops when:

•There's been no door activity from her in 18 hours, •An unknown person shows up 3 or more times in a day, or •A package has been left outside for 6+ hours.

I kept the rules deliberately boring so a family can read them and agree with them. I didn't want a mystery number. There's also an "ask about visits" box for longer questions. It pulls out the events that matter and suggests one thing to do next, usually something like "call her now and ask if she knows the visitor."

All of this is exposed as an MCP server with three tools: check_on_resident, get_recent_events and notify_family. That's the part Alexa+ actually talks to.

How we built it

•TanStack Start + React for the web app •MCP server over Streamable HTTP (spec 2025-11-25) at /mcp •An AI model for the spoken answers and the daily summary •Browser speech APIs for voice in and out, so the simulated Alexa+ actually feels like talking to something

The UI and the MCP tools read from the same data and rules module. Early on they gave slightly different answers to the same question, and that bugged me enough to restructure it.

The household is fake (Kamala, 78, and her kids Priya and Arjun). I didn't have a Ring device, and I wasn't going to point this at a real person's front door for a demo.

Challenges we ran into

Getting answers short enough to say out loud without them turning vague. My first versions read like an incident report.

Picking thresholds. 12 hours of no activity flagged every normal night's sleep. 24 felt too late. I landed on 18.

Making the voice chat feel like a conversation instead of a form. It took a few passes before it sounded like something you'd actually say to Alexa.

Accomplishments that we're proud of

The MCP server works and you can hit it right now — no login, no setup. And the alerts are explainable: every warning points back to specific events, so you can see why it's worried.

What we learned

Wiring up MCP was the easy part. The hard part was figuring out what a worried daughter actually needs to hear in ten seconds, and what she doesn't.

I also learned that a demo is only as honest as its data. Once the UI and the tools started reading from the same place, the whole thing got simpler and easier to trust.

What's next for DoorSense

•Real Ring events through Ring's APIs and simulator •notify_family sending actual texts or Alexa announcements instead of a simulated message •Learning each household's normal routine instead of fixed thresholds •A consent flow for the parent. They should be in on this, not just monitored.

⚙️ Note to Judges: YouTube defaults to dynamic "Auto" quality. To clearly read our live Model Context Protocol (MCP) server terminal logging events, please click the Gear Icon (Settings) at the bottom right corner of the embedded video player window and manually toggle the playback resolution to 1080p HD or the highest video quality.

Built With

Share this project:

Updates