CareLoop

Self-hosted Model Context Protocol (MCP) Server for Alexa+ Elderly Care Monitoring
Built for the Amazon Developer Hackathon: Alexa+ Track, AWS Builder Track, and Open Source Mini-Challenges

What Was Built During the Hackathon

Every day, millions of aging parents live independently while their adult children worry from afar. Traditional telecare systems suffer from three critical flaws:

1. Intrusive surveillance (cameras, microphones, and persistent audio recording that violate dignity).

2. Alarm fatigue (rigid static thresholds that spam family whenever Mom sleeps in 20 minutes late).

3. The AI Safety Paradox (deploying non-deterministic LLMs to make life-critical emergency alert decisions).

CareLoop solves this through a self-hosted Model Context Protocol (MCP) server over Streamable HTTP (spec 2025-11-25+) that enables Alexa+ to look after an elderly parent living alone. It passively learns each individual's personal circadian routine from ambient event metadata (pillbox openings, meals, room movement, door contacts), detects schedule deviations with circular statistics, conducts natural spoken check-ins by voice, and escalates to family caregivers via Amazon SNS only when truly necessary.

Hackathon Challenge Tracks Addressed

• Alexa+ Track:

  • Implements the official FastMCP / MCPServer Streamable HTTP transport (/mcp and /v1/mcp) with bi-directional JSON-RPC 2.0 streaming.

  • Registers 9 Alexa+ voice tools with prompt-engineered voice triggers and return schemas formatted for Alexa Text-To-Speech (TTS) synthesis.

  • Built an ambient Voice Simulator UI (simulator/index.html) featuring native browser Web Speech API microphone dictation, SpeechSynthesis audio, a live MCP inspector showing JSON-RPC packets in real time, and a 6-step automated "Dad skips lunch" demonstration.

• AWS Builder Track:

  • Amazon DynamoDB: Single-table design (PK, SK) with AWS-managed server-side encryption at rest (SSESpecification: { SSEEnabled: true }) and Point-in-Time Recovery (PITR).

  • Amazon SNS: Caregiver alert topic encrypted with AWS KMS (alias/aws/sns).

  • Amazon Bedrock (Claude 3 Haiku Converse API): Sandboxed strictly to empathetic daily routine summaries and factual memory Q&A—the LLM is strictly prohibited from making alert decisions.

  • Amazon EventBridge: Scheduled 1-minute autonomous health tick rule triggering a dedicated Lambda runner.

  • Full IaC Packaging: Production-grade AWS CDK (Python) in deploy/cdk/ and AWS SAM in template.yaml.

• Open Source Mini-Challenge:

  • 100% open-source under the MIT License.

  • Fully typed with Pydantic v2 domain models and validated by 79 pytest tests (100% pass rate).

  • Zero proprietary vendor lock-in: runs identically on local SQLite or AWS DynamoDB.

Related MCP server: earshot

Architecture

CareLoop strictly separates deterministic safety logic from generative language synthesis:

flowchart TD
    subgraph AmbientInterface ["1. Ambient Voice & Smart Home Interface"]
        Voice[("Elder Voice Input<br/>'I took my morning pills'")] --> Alexa[Alexa+ Ambient Assistant]
        Sensors[("Home Sensors<br/>Pillbox / Motion / Door")] --> Alexa
    end

    subgraph MCPGateway ["2. Streamable HTTP MCP Gateway (FastMCP)"]
        Alexa -->|"Streamable HTTP (2025-11-25)<br/>POST /mcp"| MCP["CareLoop MCP Server<br/>(Starlette ASGI on ECS Fargate)"]
        Auth["AlexaAuthMiddleware<br/>Bearer / API Key"] -.-> MCP
    end

    subgraph DeterministicEngine ["3. Deterministic Clinical Safety Engine (Zero LLM)"]
        MCP -->|"log_routine_event"| EWMA["Circular EWMA Baseline<br/>• Time-of-Day [0, 1440m)<br/>• Variance EWMA<br/>• Grace Window: μ + max(2σ, 30m)"]
        MCP -->|"check_routine_health<br/>tick_escalation"| Risk["Multi-Factor Risk Engine<br/>• Medication missed (0.4)<br/>• Inactivity / Falls (0.3)<br/>• Night wandering (0.2)<br/>• Meal missed (0.2)"]
        Risk --> StateMachine["Persisted Escalation State Machine<br/>Normal ➔ Gentle Check-In ➔ Repeat Check-In ➔ Notify Family"]
    end

    subgraph GenerativeLayer ["4. Sandboxed Generative Layer (Amazon Bedrock)"]
        MCP -->|"daily_summary<br/>ask_memory"| Bedrock["Amazon Bedrock (Converse API)<br/>Claude 3 Haiku<br/>• Warm family summaries<br/>• Factual event history Q&A<br/>(Short-circuits if no events found)"]
    end

    subgraph StorageLayer ["5. Pluggable Encrypted Storage"]
        EWMA & Risk & StateMachine & Bedrock --> Storage["Storage Interface<br/>(careloop/storage/base.py)"]
        Storage -->|"Local Dev"| SQLite[("Local SQLite DB<br/>careloop_local.db")]
        Storage -->|"AWS Production"| DynamoDB[("Amazon DynamoDB<br/>CareLoopData<br/>• Single-Table PK/SK<br/>• Encrypted at rest (KMS)")]
    end

    subgraph FamilyAlerts ["6. Caregiver Notification System"]
        StateMachine -->|"Notify Family"| Notifier["Alert Dispatcher"]
        Notifier -->|"Local Mode"| Console["Console / Terminal Log"]
        Notifier -->|"AWS Mode"| SNS["Amazon SNS Topic<br/>CareLoopAlertsTopic<br/>• SMS / Push to Daughter Linda<br/>• Deterministic Reasons Included"]
    end

Quickstart

1. Installation

# Clone the repository
git clone https://github.com/KarnailSingh87/careloop.git
cd careloop

# Create and activate Python 3.11+ virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies in editable mode with development tools
pip install -e ".[dev]"

2. Generate 60-Day Routine Benchmark Dataset

Seed the local SQLite database with 60 simulated days across 3 distinct elder personas (eleanor, dad, martha), yielding 1,422 routine events, 18 ground-truth safety incidents, and 42 benign noise controls:

python scripts/seed_data.py

3. Launch Local MCP Server

Start the FastMCP Streamable HTTP server with the simulated clock enabled:

# Using module launcher
python -m careloop.server --host 127.0.0.1 --port 8000 --clock simulated

# Or using the root entrypoint
python server/main.py --host 127.0.0.1 --port 8000 --clock simulated

• MCP Streamable HTTP Endpoint: http://127.0.0.1:8000/mcp

• Web Simulator UI & MCP Inspector: http://127.0.0.1:8000/simulator

• Health Check Endpoint: http://127.0.0.1:8000/health

4. Test Streamable HTTP MCP Tools

Run the included integration client to verify session negotiation, tool discovery, and execution over Streamable HTTP:

python scripts/client_test.py --url http://127.0.0.1:8000/mcp

5. Run Full Test Suite

pytest -v
# 79 passed in 6.17s

Web Simulator & Live Demo

CareLoop includes an interactive single-page application at http://127.0.0.1:8000/simulator styled after an ambient Alexa+ smart display:

• Ambient Alexa Orb: Animated glowing orb reflecting idle, listening, thinking, and speaking states.

• Voice Microphone (STT): Uses the browser Web Speech API to dictate natural routine statements directly to Alexa+.

• Speech Synthesis (TTS): Reads Alexa voice responses aloud with audio chimes.

• Live MCP Tool Inspector: Displays exact JSON-RPC 2.0 tool requests, input arguments, execution latency, and return schemas (2025-11-25).

• Injectable Clock Machine: Jump forward in time (+15m, +1h, +1d, Jump to 2:00 PM) to test missed routines without waiting hours.

• Virtual Family Phone: Displays push/SMS notifications received by family caregivers with deterministic reasons and resolution status.

• "Dad Skips Lunch" Hero Demo: One-click 6-step live demonstration that plays the full lifecycle:

  1. Morning: Medication and breakfast logged normally.

  2. 12:45 PM: Normal lunch window arrives. Dad takes a long nap.

  3. 01:45 PM: Grace window expires; risk elevated to Medium; Alexa initiates a gentle voice check-in.

  4. 01:47 PM: No response; repeated check-in with attention chime.

  5. 01:49 PM: State machine escalates to notify_family; alert dispatched via Amazon SNS with audit reasons.

  6. 02:00 PM: Dad wakes up and says "I'm fine"; state machine resets to normal and family phone resolves.

Alexa+ MCP Tool Reference

CareLoop exposes 9 specialized MCP tools designed for Alexa+ natural voice interaction:

Clinical Benchmark & Evaluation Results

CareLoop was benchmarked across a synthetic dataset of 60 simulated days per persona (180 elder-days total) across 3 clinical personas with realistic Gaussian jitter and 18 injected safety incidents:

To reproduce the benchmark evaluation:

python scripts/evaluate_benchmark.py

Strict Privacy Architecture

CareLoop is built from the ground up for strict healthcare and elder privacy:

1. Zero Raw Audio Storage: Neither CareLoop nor the MCP server ever captures, records, or stores ambient audio. Only discrete metadata tuples are stored: (person_id, event_type, timestamp, metadata).

2. Deterministic Safety Isolation: Amazon Bedrock is used strictly for phrasing daily summaries and answering factual routine questions. The LLM cannot invoke emergency alerts or change escalation stages.

3. Auditability & Explainability: Every alert dispatched to family contains deterministic, explainable reasons (e.g. "Blood pressure medication window closed at 09:15 UTC. Routine check-in unanswered after 120 seconds.").

4. Encryption at Rest: On AWS, all DynamoDB items and Amazon SNS messages are encrypted with AWS KMS customer-managed keys.

AWS Cloud Deployment

Deploy CareLoop to AWS with either AWS CDK or AWS SAM:

Option A: AWS CDK (Python)

cd deploy/cdk
pip install -r requirements.txt
cdk bootstrap aws://<ACCOUNT_ID>/us-east-1
cdk deploy

Option B: AWS SAM CLI

sam build
sam deploy --guided

Outputs Provisioned

• Public MCP Streamable HTTP Endpoint: http://<ALB_DNS>/mcp

• Web Simulator UI: http://<ALB_DNS>/simulator

• Health Check: http://<ALB_DNS>/health

• Encrypted DynamoDB Table: CareLoopData

• Amazon SNS Topic: CareLoopAlertsTopic

• Amazon EventBridge: 1-minute automated tick rule calling CareLoopTickRunner Lambda.

Open Source & MIT License

CareLoop is free and open-source software licensed under the MIT License.

Copyright (c) 2026 CareLoop Contributors
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction...