Inspiration

Millions of families worry about elderly parents or loved ones who live alone. A missed phone call or an unusual change in daily activity can leave family members wondering whether everything is okay. However, installing cameras inside someone's home can compromise their privacy and independence.

We wanted to create a solution that helps families stay informed without constantly monitoring their loved ones. This inspired us to build Checkin, a privacy-conscious, AI-powered home activity assistant that helps families understand whether someone's daily routine appears normal or unusual.

Instead of relying on rigid alarms or fixed schedules, Checkin learns an individual's typical activity patterns over time and identifies meaningful changes. Our goal is to provide peace of mind to families while respecting the dignity, independence, and privacy of the person living alone.

What it does

Checkin helps families understand the daily activity patterns of a loved one who lives alone, without requiring cameras inside the home.

Users can ask Alexa+, "Is Mom okay today?" Checkin analyzes simulated home sensor events and compares the person's current activities with their usual routine over the previous 14 days.

For example, Checkin might report that kitchen activity started approximately 25 minutes later than usual, no medicine cabinet activity has been detected yet, and the front door has not opened in two days.

Its key features include:

• Personalized Routine Monitoring: Compares current activity with the individual's historical patterns rather than relying on generic alarm rules.
• Daily Activity Insights: Tracks simulated activity around important areas, including the kitchen, front door, mailbox, and medicine cabinet.
• Routine Deviation Detection: Highlights unusual changes in activity timing or frequency that may deserve attention.
• Natural-Language Responses: Presents activity insights in a conversational format, making information easier for family members to understand.
• Family Check-ins: Provides a tool to log check-ins and maintain a record of family interactions.
• Message Drafting: Helps prepare messages to family members when a follow-up may be appropriate.
• Privacy-Focused Design: Uses activity-event data instead of indoor camera footage.

Checkin is designed to help families recognize potential concerns earlier. It identifies unusual activity patterns, but it does not independently confirm someone's health or safety.

How we built it

We built Checkin using a Model Context Protocol (MCP) architecture that connects a simulated home-monitoring environment with a simulated Alexa+ experience.

The project consists of three main components:

1. Self-Hosted MCP Server

We developed a real MCP server using the TypeScript SDK and Streamable HTTP transport. It exposes four tools that allow an AI assistant to retrieve activity information, compare routines, record check-ins, and prepare family messages.

The four tools are:

• get_today_rhythm — Retrieves the person's activity pattern for the current day.
• compare_to_usual — Compares current activity against the person's historical routine.
• log_check_in — Records a family check-in.
• draft_message_to_family — Drafts a message to family members when follow-up may be needed.

2. Simulated Home Sensor Data

To demonstrate the system, we created simulated Ring-style sensor events representing 14 days of normal activity and one unusual day.

These events represent activities associated with different areas of the home, such as the kitchen, front door, mailbox, and medicine cabinet.

The historical data establishes a baseline for the person's routine, allowing the system to identify deviations and generate contextual insights.

3. Simulated Alexa+ Experience

We created a web-based chat interface and a lightweight backend that acts as an MCP client.

When a user asks a question, the backend communicates with the actual MCP server through the MCP protocol, invokes the appropriate tools, and uses the returned information to provide a conversational response.

Although the sensor data and Alexa+ experience are simulated, the MCP server is real, and the simulator communicates with it through the actual protocol.

The project is open source and released under the MIT license, allowing developers to explore, adapt, and extend the implementation.

Challenges we ran into

One of our main challenges was designing a system that could recognize unusual activity without depending on rigid rules. People's daily routines naturally vary, so a fixed alarm time may generate unnecessary alerts or overlook meaningful changes.

We addressed this challenge by using a 14-day historical activity baseline and comparing current events against the individual's own routine.

Another challenge was integrating the simulated home environment with the MCP server. We needed to ensure that the web interface and backend could invoke the appropriate tools through the MCP protocol rather than simply displaying predefined responses.

We also had to make activity insights understandable and useful. Raw sensor events are not meaningful to most family members, so the system needed to transform activity information into clear, natural-language explanations.

Privacy was another important consideration. Our goal was to demonstrate useful home activity monitoring without relying on indoor camera footage. Using simulated sensor events allowed us to demonstrate the concept while keeping the project focused on privacy-conscious monitoring.

Finally, we needed to communicate the limitations of the prototype clearly. Unusual activity can indicate a possible concern, but the absence of an event does not prove that someone is in danger. Checkin therefore presents activity patterns as contextual information rather than definitive health or safety assessments.

Accomplishments that we're proud of

We are proud to have built a working MCP-based prototype that demonstrates how AI assistants can interact with structured home activity data through real tool calls.

Our key accomplishments include:

• A functional MCP server: Implemented four dedicated tools using the TypeScript SDK and Streamable HTTP transport.
• Real protocol integration: Connected the simulated frontend and backend to the MCP server through the MCP protocol.
• Personalized activity comparison: Demonstrated how a person's current routine can be compared with 14 days of historical activity.
• Context-aware insights: Generated natural-language explanations of activity deviations instead of relying solely on fixed alarms.
• Privacy-conscious monitoring: Demonstrated a camera-free approach using simulated sensor events.
• Interactive user experience: Created a web-based chat interface that allows users to ask questions about daily activity.
• Open-source contribution: Released the project under the MIT license to support experimentation and further development.

Most importantly, we demonstrated how an AI assistant can turn simple activity events into information that may help families make more informed decisions about checking on loved ones.

What we learned

Building Checkin helped us understand how AI applications can move beyond simple question-and-answer interactions by using tools to retrieve and process structured information.

We learned how the Model Context Protocol enables an AI-facing application to communicate with external capabilities through a defined interface. Implementing a real MCP server also helped us understand the importance of clear tool definitions, backend integration, and reliable communication between components.

We also learned that personalization matters when interpreting human behavior. A routine that is normal for one person may be unusual for another, making individual historical patterns more useful than universal thresholds.

Another important lesson was that privacy and usefulness do not have to be opposing goals. Activity-based monitoring offers a way to explore home safety applications without requiring continuous indoor video surveillance.

Finally, we learned the importance of distinguishing between observed data and inferred conclusions. A system should explain what it has detected, acknowledge uncertainty, and avoid claiming that a person is safe or unsafe based solely on sensor events.

These lessons will guide our future work toward building AI systems that are practical, transparent, privacy-conscious, and genuinely useful to people.

What's next for Checkin

Our current prototype demonstrates the core concept using simulated sensor events and a simulated Alexa+ interface. Our next goal is to move toward a more realistic, reliable, and accessible home activity assistant.

Potential future improvements include:

• Real Sensor Integration: Connect compatible door, motion, and appliance sensors to replace simulated activity events.
• Improved Routine Learning: Develop more robust methods for learning individual routines and accounting for weekends, holidays, and natural variations.
• Configurable Privacy Controls: Give individuals greater control over which activities are collected, how long data is retained, and which family members can access it.
• Consent-Based Family Sharing: Introduce clear consent and permission controls before sharing activity information with relatives.
• Timely Notifications: Provide configurable notifications when significant deviations occur, with safeguards against unnecessary alerts.
• Voice Assistant Integration: Explore integration with supported voice assistant platforms instead of relying on a simulated Alexa+ experience.
• Reliable Check-in Workflows: Improve family check-in records and message drafting so that potential concerns can be followed up on appropriately.
• Real-World Testing: Evaluate the system with consent-based testing, realistic activity scenarios, and feedback from families and people living independently.
• Security and Reliability: Strengthen authentication, data protection, error handling, and monitoring before any real-world deployment.

Our long-term vision is to make Checkin a trustworthy companion for independent living: a system that helps families stay connected and informed while allowing their loved ones to maintain their privacy, freedom, and dignity.

Checkin is built around one simple belief: caring for someone should not mean watching their every move.

Built With

• express.js
• model-context-protocol-(mcp)
• node.js
• streamable-http
• typescript