Clinic Access Navigator

Inspiration

As a physician and founder working with private clinics, I have seen how difficult it can be for patients to navigate even the first step of care. They may not know which service to choose, which branch to visit, when an appointment is available, or how to prepare.

Clinic staff repeatedly answer these questions through phone calls and messaging apps. The problem is not always a lack of medical care—it is often the friction involved in accessing it.

Clinic Access Navigator explores how Alexa+ and the Model Context Protocol can make that journey easier while maintaining a clear boundary between patient access and medical diagnosis.

What it does

Clinic Access Navigator helps patients find clinic information and complete a demonstration appointment booking through a guided conversation.

The updated prototype remembers booking choices during the conversation, displays available slots, and asks for explicit confirmation before saving an appointment. Users can then view their demo appointments and cancel a reservation with confirmation.

The demonstration includes two fictional ENT doctors, scheduled availability, and synthetic adult patient profiles. Reservations are stored on the server, with safeguards against duplicate bookings and occupied slots.

The application communicates with an MCP server through actual tool requests. Its conversational interface currently uses rule-based logic. Alexa+ device integration, voice input, and Amazon Bedrock inference are not connected in this version.

All appointments and patient profiles are fictional. The prototype provides administrative assistance, not medical diagnosis or treatment.

How we built it

Clinic Access Navigator is a web application with a guided conversational interface and an MCP server. The current conversation flow uses rule-based logic to collect booking details and retain context during the active conversation.

The application performs the MCP initialization, tool discovery, and tool-call lifecycle. Tools support clinic information, visit preparation, booking options, available slots, appointment creation, appointment retrieval, and cancellation.

The booking tools include:

• get_demo_booking_options
• list_demo_slots
• book_demo_appointment
• list_demo_appointments
• cancel_demo_appointment

The demonstration uses two fictional doctors: ENT Doctor A, available Mondays to Fridays from 11 AM to 3 PM, and ENT Doctor B, available Tuesdays to Saturdays from 4 PM to 7 PM. Appointment times use the Philippine time zone.

Demo reservations are stored on the server. Booking safeguards prevent occupied slots from being booked again and protect against duplicate requests. Users review their details and explicitly confirm before creating or cancelling an appointment.

Only fictional adult patient profiles are used. Alexa+ device integration, voice input, Amazon Bedrock inference, and Google Calendar synchronization are not connected in this version.

I defined the clinic workflows, scope, and healthcare boundaries, and used ChatGPT/Codex to assist with implementation, testing, documentation, and demonstration preparation.

Challenges we ran into

A central challenge was keeping the assistant focused on administrative tasks. The prototype helps users navigate clinic information and appointments without diagnosing conditions or recommending treatment.

Another challenge was maintaining context across follow-up messages while ensuring that booking and cancellation require explicit confirmation. Server-side validation and persistent reservations were needed to make these actions reliable.

We also encountered AWS account-verification and model-quota restrictions while testing Amazon Bedrock. A successful Bedrock connection has not been established, so the current demonstration does not depend on it.

Accomplishments that we're proud of

The project translates experience operating private clinics into a working demonstration of conversational patient access.

Users can move from an initial booking request through doctor and slot selection, fictional patient selection, review, and confirmation. They can then retrieve their demo appointments and cancel a reservation.

The prototype makes actual MCP tool calls and saves demo reservations on the server. The updated screen recording demonstrates the booking, retrieval, and cancellation flow using the local application and MCP server.

What we learned

Conversational booking requires more than a friendly interface. It needs clear state management, validated availability, reliable persistence, and explicit confirmation before changing appointments.

MCP tools give each administrative action a defined purpose, structured inputs, and inspectable results. Separating these tools from the conversational interface makes the workflow easier to test and extend.

We also learned to distinguish implemented functionality from planned integrations. Voice accessibility remains a future goal that requires implementation and user testing.

What's next for Clinic Access Navigator

Connect and validate the experience with Alexa+.

Resolve AWS access restrictions and evaluate Amazon Bedrock for language understanding.

Deploy and validate the MCP service on AWS.

Add direct rescheduling with safeguards that preserve the original appointment if the replacement cannot be confirmed.

Expand accessibility support, including voice interaction.

Test additional synthetic clinic scenarios and conduct structured usability testing.

Add appropriate authentication, privacy controls, and operational safeguards before considering real patient use.

The long-term goal is to help private clinics offer easier conversational access alongside their existing clinical and operational systems.

Built With

• ai
• alexa+
• application
• context
• conversational
• data
• healthcare
• interface
• model
• protocol
• synthetic
• technology
• user
• voice
• web