Context-Aware Personal Assistant MCP Server for Alexa+

Persistent memory, tasks & calendar for Alexa+ via MCP 2025-11-25 Streamable HTTP

Comment

Context-Aware Personal Assistant MCP Server for Alexa+ & Nebius

Problem

AI assistants lack persistent memory across conversations. Every session starts fresh. Users can't build long-term relationships with their assistant.

Solution

A Model Context Protocol (MCP) server implementing the 2025-11-25 Streamable HTTP specification that gives AI assistants:

  • Persistent Memory - Store/retrieve memories with tags, importance scoring, semantic search
  • Task Management - Create, update, track tasks with priorities, due dates, tags
  • Calendar Integration - Schedule events, get upcoming events, manage recurrence
  • Proactive Assistance - Context-aware suggestions based on time, events, tasks
  • User Preferences - Persistent preference storage for personalization

Nebius + NVIDIA Integration

  • Runs on Nebius Token Factory - Uses Nemotron 3 Ultra for reasoning, Nemotron Nano/Super for fast responses
  • Nebius Serverless Endpoints - Deploys the MCP server for scalable inference
  • Nebius Serverless Jobs - Background processing for memory consolidation, proactive suggestion generation
  • NVIDIA Nemotron Models - Core LLM powering the assistant's intelligence
  • Hermes Agent - Local agent framework for personal AI assembly

Architecture

User → Alexa+/Client → MCP HTTP Server (Nebius Serverless) → Nemotron (Nebius Token Factory) → Persistent JSON Storage

Technical Highlights

  • Full MCP 2025-11-25 Compliance: Streamable HTTP + SSE, proper initialization, session management
  • 11 Tools: add_memory, search_memories, get_recent_memories, create_task, update_task, list_tasks, add_event, get_upcoming_events, get_context, get_proactive_suggestions, set_preference
  • 5 Resources: assistant://context, assistant://memories, assistant://tasks, assistant://events, assistant://preferences
  • Production Ready: Docker multi-stage build, health checks, non-root user, CORS for Alexa+
  • Privacy-First: Local JSON storage, no external dependencies beyond Nebius/NVIDIA

Track: Personal AI

Building an always-on, private assistant that works for you while keeping your data under your control. Uses Nemotron models on Nebius Token Factory and tools like Hermes Agent to assemble, secure, and run your own personal AI system.

Demo

[Video demo to be uploaded to YouTube]

GitHub Repository

https://github.com/ASAYAMedia/amazon-alexa-mcp-hackathon

Built With

  • Python 3.11
  • MCP Python SDK
  • FastAPI / Uvicorn
  • Pydantic
  • httpx
  • Docker
  • Nebius Token Factory (Nemotron 3 Ultra, Nano, Super)
  • Nebius Serverless Endpoints
  • Nebius Serverless Jobs
  • Hermes Agent

What I Learned

  • MCP Streamable HTTP transport implementation details
  • Session management patterns for MCP
  • Building privacy-first local-first applications
  • Alexa+ integration requirements
  • Nebius Token Factory + Nemotron model integration
  • Serverless AI deployment patterns

Challenges

  • Implementing custom HTTP transport (MCP SDK only provides STDIO)
  • Session management with Mcp-Session-Id header
  • SSE endpoint for real-time updates
  • Integrating Nemotron models via Nebius Token Factory API
  • Optimizing model selection (Ultra vs Nano/Super) for cost/performance

Built With

Share this project:

Updates

Submission history