About the project

Inspiration

Ever since I was a kid, I've had family members who are doctors, and they have always inspired me to be at the service of others. But something else I'm passionate about is technology: I've always loved building projects around it, and that's what steered me toward building technology for healthcare. Every day I see doctors who hate having to write on paper or type into a computer, losing time when what they really want is to focus on giving the patient more human care.

Honestly, we had already been thinking about the huge possibility here. There are apps that can fill in fields through voice dictation, but they're not that fast and their accuracy isn't exactly the best. So when we saw how well this adapts through WebMCP, it clicked: it's literally a perfect symbiosis between human and AI.

What it does

It basically creates the possibility for the doctor to control their whole system through chat or voice: checking a patient's status, signing consultations, and more, freeing up their time so they can be more present with the patient.

Why this use case is a strong fit for WebMCP

The agent takes only what it is allowed to take, and it produces faster, more precise results.

No hospital hands its EHR credentials to an external agent. That single fact kills every classic integration path (API contracts, OAuth scopes, security audits, etc.). With WebMCP there is no second party: the agent works inside the session the doctor already opened, in the same tab, with the same permissions, and every action is visible on the very screen the doctor is looking at. Tools are context-scoped, so the agent can only ever do what the current screen can do. Remove WebMCP and this isn't less convenient: it's unshippable.

What people and agents can do together

The doctor's electronic health record stops being a burden and becomes their personal assistant. But something the system can NEVER do is take actions that directly affect the patient WITHOUT the doctor's authorization.

How we built it

It grew out of inspiration from several electronic health records, redesigned to be more streamlined for the user, and then adapted for the agent.

Under the hood it's React 18 + Vite: a clean-room slice of a production chart UI running on a mocked data layer with 100% synthetic patients, so the live demo needs no backend and zero API keys. WebMCP is the imperative API behind feature detection: document.modelContext.registerTool(toolDef, { signal }) with one AbortController per UI scope. The patient list registers 3 tools, opening a chart swaps them for 7 chart-scoped tools, and closing it unregisters them (toolchange in action). Read tools carry readOnlyHint, and the only write tool creates a DRAFT that the doctor must review and sign in the UI. The agent cannot sign.

What surprised us

The system now literally becomes the doctor's best friend, reading years of medical history within a 15-minute visit, something that used to be really hard.

What's next

My takeaway is that this is just the tip of the iceberg. This could also be implemented in an imaging viewer for X-ray and MRI, where the doctor controls a full 3D model of the body and analyzes it like Ironman. We'll keep exploring what's possible.

Built With

• chrome
• claude
• framer-motion
• gemini
• i18next
• json-schema
• jspdf
• material-ui
• model-context-protocol
• netlify
• react
• react-query
• recharts
• typescript
• vite
• webmcp