CareLoop

Inspiration

Many of us live far from our parents. The daily questions are simple but stressful: did Mom take her morning pills, and how is Dad feeling? Answers are scattered across phone calls and family group chats. Caregivers are usually mid-task, in the kitchen or driving, so voice is a natural fit.

What it does

CareLoop is an Alexa+ caretaking assistant. A family caregiver can speak naturally:

"I gave Mom her metformin" logs the dose against her schedule. "Which doses were missed?" reports overdue medications across everyone in the care circle. "Mom's mood is 2 today" records a check-in. Low moods automatically alert that person's caregivers. "Dad felt dizzy this morning" raises a concern and notifies the family. "Give me the daily summary" returns a warm, plain-language digest per person.

It handles several people, each with their own medications and caregivers. It infers who you mean from the medication ("amlodipine" means Dad), and asks "Whom do you mean?" when a request is ambiguous. When you report a dose, it catches up the earliest overdue one first, so a late evening report doesn't hide a missed morning dose.

A live family dashboard shows every dose's status (taken, due, missed, upcoming) and the concerns that were sent.

How we built it Alexa+ integration: CareLoop is a self-hosted MCP server (Streamable HTTP, /mcp) built with the official TypeScript MCP SDK, exposing seven tools: list_people, get_care_schedule, log_medication, get_missed_doses, record_checkin, flag_concern, and daily_summary. Simulated Alexa+ experience: We had no Alexa+ device, so we built a web app that simulates the voice experience. It supports typing or the browser microphone, speaks replies, and shows each tools/call it makes. It speaks to the real MCP server over JSON-RPC, so what you see is the same protocol an Alexa+ client would use. Backend: Node.js, Express and TypeScript. The dose-status logic covers a grace window, timezone-aware schedules, and slot catch-up. Storage is a JSON file, and a DynamoDB backend is included for hosting. Testing: A smoke test connects with a real MCP client over HTTP and exercises every tool, including ambiguity handling and per-person alerts. A fixed-clock test covers the slot-picking rule. Digest: The daily digest can be generated by an LLM through Amazon Bedrock's Converse API when configured (untested here). Otherwise it uses a deterministic template. We did not use Bedrock for this submission, so we aren't entering the AWS Builder challenge. Deployment: A live demo runs at [your Render URL]. Challenges Making natural phrasing map cleanly to tools: "her", "mom", and a bare medication name all need to resolve to the right person without guessing wrongly on health-related actions. We chose to ask a clarifying question over guessing. Deciding which slot "I gave her metformin" refers to when doses are missed, which led to the catch-up rule. Keeping the MCP server stateless so it could run on serverless platforms. Accomplishments we're proud of

A complete voice-to-alert loop over a standards-based MCP server, with tests, plus a clear demo that shows the protocol calls happening.

What we learned

MCP's tool-descriptions-as-interface model works well for voice. Good tool descriptions and optional arguments matter more than complex routing.

What's next

Real notifications (SMS or push) to caregivers, per-user accounts with authentication, medication schedule editing by voice, and running against a real Alexa+ device once available.

Built with

TypeScript, Node.js, Express, Model Context Protocol (Streamable HTTP), Zod, Amazon DynamoDB and Lambda (deploy-ready, untested), Amazon Bedrock (optional), Render

Open source

MIT licence: https://github.com/Ayush-web28/CareLoop

(CareLoop tracks and relays information. It is not a medical device and gives no medical advice.)

Tracks and challenges Track: Alexa+ (simulated-experience path, backed by a real MCP server) Mini challenge: Open Source Existing project? No. It was built during the hackathon window. Open Source challenge fields Repo URL: https://github.com/Ayush-web28/CareLoop GitHub username: Ayush-web28 Contribution URL: [same repo URL, unless you make a separate PR or fork] Description: A new open-source (MIT) MCP server for caretaking, with a simulated Alexa+ client and tests. It shows how a small, self-hosted MCP server can give Alexa+ a real-world caregiving capability.