PERI for Alexa+

A hot flash doesn't wait for you to unlock your phone. PERI's MCP server lets women log symptoms and talk to Peri, their perimenopause coach, by voice.

Comment

Inspiration

More than a billion women worldwide are in or past menopause, and the years leading up to it (perimenopause) bring hot flashes, night sweats, brain fog, and 3 AM wake-ups that many women are told to just push through. My fiancé Nicky and I built PERI, a perimenopause wellness app now live on the App Store, so women have a place to track what's happening and a coach, Peri, who actually listens.

But we kept hearing the same thing: symptoms happen at the worst moments. In the middle of cooking, in a meeting, in bed at 3 AM. Nobody wants to unlock a phone, find an app, and tap through a form while they're overheating. Voice is the natural way in, and Alexa+ is where people already talk to their homes.

What it does

PERI for Alexa+ lets a woman talk to PERI out loud:

  • "I'm having a hot flash." PERI logs it in her tracker, then Peri talks her through it in about twenty seconds: cool your wrists and neck, loosen a layer, slow breathing in for five and out for five, with the source named (The Menopause Society). If it's her third this week, Peri gently mentions the pattern.
  • "How has my week been?" A spoken summary of what she's tracked and what stands out.
  • "Ask Peri why I keep waking up at 3 AM." A full answer from Peri, drawing on her real history in the app, saved to her Peri chat so she can pick it up later on her phone.
  • "Find me a video to help me sleep." Suggests one of PERI's guided breathing or movement videos.
  • "My period started today, it's light." Marks her cycle calendar.

Everything said by voice shows up in the PERI app instantly.

How we built it

The heart of the project is PERI's MCP server, built with the official Model Context Protocol TypeScript SDK (1.32.1) using the Web Standard Streamable HTTP transport (protocol 2025-11-25), running statelessly as a Supabase Edge Function next to PERI's existing backend. It exposes five tools: log_symptoms, log_period_day, get_symptom_summary, find_videos, and ask_peri.

Security was the first design decision, because this is health data. The caller passes the member's own PERI access token, and every query runs as her under the same Postgres row-level security as the app, so the server can only ever see or change her own data. Unauthenticated requests are rejected before any tool runs. ask_peri forwards to PERI's existing coach, so Peri's usage limits and safety rules (never diagnose, never recommend prescriptions, emergencies go to 911) carry over to voice automatically.

Because Alexa+ is in limited preview, we followed the hackathon's guidance and built a simulated Alexa+ experience: a web page with speech recognition and a calm female voice, backed by an agent (Claude, through Anthropic's MCP connector) that reaches PERI only through the MCP server, the same way Alexa+ consumes an MCP integration. Each reply shows which MCP tool ran, so judges can see exactly what happened. On real Alexa+, the simulator drops away and Alexa+ connects to the same server.

Challenges we ran into

  • Making voice answers useful, not just short. Our first version said "Logged your hot flash, try a few deep breaths," which felt hollow next to how Peri talks in the app. We rewrote the voice persona to speak as Peri, give two or three concrete steps with a source, then tuned it down from about 45 seconds of speech to about 20, because someone mid hot flash doesn't want a lecture.
  • Account linking for health data. We needed a way for a voice assistant to act for one specific woman without ever touching anyone else's data. Passing her own token through and letting the database's row-level security enforce it kept the MCP server simple and safe.
  • Testing without Alexa+ access. Without preview access we couldn't test on a real Alexa+ device, so we built the simulator to exercise the exact same MCP endpoint.

Accomplishments that we're proud of

  • All five tools work end to end on PERI's live production backend, against real accounts.
  • Peri's answers by voice carry her memory: asked about 3 AM wake-ups, she brought up the user's frequent travel from earlier conversations in the app.
  • Zero changes to the shipped mobile app were needed. The MCP server sits beside it and reuses its database, security, and coach.

What we learned

  • MCP turns an existing app into something any assistant can use: one well-described set of tools and the assistant figures out the rest, including mapping "I'm burning up" to a hot flash.
  • Voice needs its own writing style. Lists and formatting don't exist out loud, and length matters far more than on a screen.

What's next for PERI for Alexa+

  • Connect to real Alexa+ as soon as preview access allows, with proper account linking (OAuth) so a woman can link PERI from the Alexa app.
  • Proactive check-ins: "Good morning, how did you sleep?" feeding Peri's daily check-in.
  • Spoken guided breathing sessions during a hot flash, and a Fire TV version of PERI's video library.

Built With

Share this project:

Updates

Submission history