Careloop

Inspiration

Many of us live far from our parents. We worry every day, but a phone call is not always honest ("I'm fine"), and doorbell or camera alerts are too noisy to check. A real problem hides inside the noise: a long quiet period, when nothing happens for hours. I wanted one short answer a family can understand in 10 seconds.

What it does

CareLoop turns simple home events (doorbell and motion) into clear answers to questions like "Is my mother okay today?", "Who visited today?" and "When was the last activity?"

• On a normal day it says: "Yes, today looks normal."
• If nothing happens for 12 hours or more during the day, it says: "Please check on your parent."

How I built it

• MCP server (Alexa+ track): a self-hosted MCP server using Streamable HTTP, built with the Python MCP SDK (FastMCP). It exposes four tools: get_today_summary, get_last_visitor, is_everything_ok and get_last_activity_time.
• Plain rules for safety: a small Python function finds the longest quiet period in the day. The safety decision is made by simple, predictable rules, not by an AI guess.
• Simulated Alexa+ experience: a Streamlit chat page that calls the same tools, with a switch between a normal day and a no-activity day.
• Testing: I tested every tool with MCP Inspector.

Challenges I ran into

• The mcp package now installs version 2.x by default, which renamed FastMCP. Sample code failed with an import error until I pinned mcp<2.
• MCP Inspector 2.9.0 changed to a server list, and its "Add server" form defaults to stdio, so connecting to an HTTP server was confusing.
• I researched the Ring API, but it needs a Ring account with a Ring Protection plan and, per the docs, only supports devices located in the US. I could not use it from outside the US, so I chose the Alexa+ track instead.

What I learned

I am a beginner. I learned Python, Git and GitHub, how MCP servers and tools work, how to test with MCP Inspector, and how to build a simple web page with Streamlit.

Important note

All home events in this project are simulated. It is not connected to a real device or a real person's home, and it is not an emergency service.

What's next

Connect real event data, add Alexa+ registration when it is available to developers, and let an AI model understand more ways of asking the same question while the rules still make the safety decision.

Built With

• alexa+
• fastmcp
• github
• mcp
• mcp-inspector
• python
• streamable-http
• streamlit