Livevault

A voice-first life-admin assistant via Streamable HTTP MCP with AWS Lambda and DynamoDB. Never miss a warranty, renewal, or deadline again.

Comment

Inspiration

Warranties, utility bills, subscriptions, and renewal dates are perpetually scattered across email inboxes, paper folders, and junk drawers. Most people only realize an item's warranty expired or a bill rate spiked long after the fact. While general-purpose voice assistants like Alexa+ excel at orchestrating conversations, they lack a private, structured repository for personal lifecycles and frequently struggle with deterministic calendar math and anomaly detection. We built LifeVault to give voice assistants a deterministic, secure memory layer for everyday life administration.

What it does

LifeVault is an open, Model Context Protocol (MCP) spec 2025-11-25 compliant server deployed over Streamable HTTP. It exposes 14 specialized life-admin tools to conversational clients (such as Alexa+ or our standalone web simulator):

  • Deterministic Record Tracking: Ingests purchases and renewals via natural speech, automatically calculating exact countdowns and 30-day proactive reminder thresholds in code rather than guessing through an LLM.
  • Proactive Voice Briefings: Delivers a daily voice briefing summarizing items due in the next 7 days, recent expirations, and utility bill spikes (>25% variance over median).
  • Automated Claims Drafting: Assembles structured claim emails directly from verified backend records, inserting placeholders where facts are unknown to prevent hallucinated details.
  • Security & Privacy Guardrails: Enforces constant-time Bearer token authentication, fail-closed cloud execution, spoken two-step confirmation on destructive actions, and regex/Luhn-based privacy filters that automatically reject credit cards, PINs, passwords, and sensitive identifiers.

How we built it

  • Protocol & Backend: Built on Python using FastMCP and FastAPI, adhering to the latest Streamable HTTP MCP specification (2025-11-25).
  • Cloud & Serverless: Deployed to AWS Lambda in ap-southeast-2 using a Mangum ASGI adapter behind a secure Lambda Function URL.
  • Persistence: Implemented a single-table design on Amazon DynamoDB (lifevault-records) with atomic counter increments for ID generation, backed by an abstract storage interface that supports local SQLite testing.
  • Pluggable AI Layer: Engineered dual-provider LLM support—defaulting to Google Gemini for zero-cost execution with an enterprise-ready Amazon Bedrock Claude-3 Haiku fallback path.
  • Testing: Authored 65 automated pytest unit and contract tests using moto.mock_aws.

Challenges we ran into

Building across bleeding-edge protocols and cloud boundaries presented concrete integration hurdles:

  1. MCP Inspector vs. Static Tokens: The official MCP Inspector attempted OAuth/DCR discovery and failed against static pre-shared Bearer tokens, prompting us to build a custom simulator.
  2. Cloud Regional Restrictions: Working around initial Free Plan regional locking in ap-southeast-2 required parameterizing the entire SDK stack.
  3. Packaging Serverless Wheels: Reconciling local Windows environments with compiled Linux wheels (manylinux2014_x86_64) for Python 3.13 Lambda runtimes.
  4. Detailed documentation of all 10 friction points encountered during development is cataloged in our repository's FRICTION_LOG.md.

Accomplishments that we're proud of

  • Zero-Hallucination Math: Delegating date arithmetic, countdowns, and utility anomaly detection entirely to deterministic Python code.
  • Production Test Coverage: Reaching 65 passing automated tests verifying storage contracts, privacy filters, tool execution, and Lambda auth handlers.
  • Open Source Contribution: Shipping the standalone Alexa+ Web Simulator as an open-source test harness to help other developers test Streamable HTTP MCP servers without OAuth headaches.

What we learned

We learned the power of the Model Context Protocol in decoupling conversational interfaces from data authority. We gained hands-on experience structuring single-table DynamoDB schemas, managing ASGI serverless adapters with Mangum, and implementing privacy filtering layers before user text ever reaches backend storage.

What's next for LifeVault

  • Adding multi-tenant user authentication via OAuth 2.0 / Amazon Cognito.
  • Field-level KMS encryption for all stored policy and receipt metadata.
  • Integrating push-notification reminders directly to Echo hardware via proactive Alexa Skills messaging.

Built With

Share this project:

Updates

Submission history