Alexa+ is powerful, but it forgets everything between conversations. alexa-memory-mcp is a self-hosted MCP server implementing MCP spec 2025-11-25 over Streamable HTTP, wrapping a knowledge-graph memory engine to give Alexa+ persistent long-term memory.
Inspiration
Every conversation with Alexa+ starts from a blank slate. I wanted her to actually remember me — my preferences, my family, my routines — across sessions.
How I built it
A real MCP server built with the MCP Python SDK (FastMCP), served over Streamable HTTP with uvicorn. Four tools:
- memory_tell — stores facts as graph triples (LLM-extracted)
- memory_ask — answers questions via planned graph lookups
- memory_forget — erases an entity
- memory_stats — reports counts
A simulated Alexa+ web demo talks to the real server over JSON-RPC. Runs on Gemini's free tier or fully offline.
Challenges
Handling Gemini API 429 bursts with backoff retries, and making sure extracted facts land as clean graph triples.
What's next
More extraction quality, richer relation types, and a real Alexa+ skill integration.
Open source, MIT license: https://github.com/Velu2k03/alexa-memory-mcp
Log in or sign up for Devpost to join the conversation.