Labsco
MCP SERVER

Jean Memory

by jonathan-politzki ↗

A memory layer with an orchestration tool in front of it — one call decides what context the model needs, gathers it, and saves new information in the background.

Agent Memory
Summary
One call that decides what to retrieve, rather than four you have to sequence.

Most memory servers hand you search and store and leave the strategy to the model — which means the model spends turns deciding whether to search, and how broadly. The orchestration tool moves that decision server-side: it reads the message and the history, picks the retrieval approach, and returns an assembled context package. The granular tools stay exposed, so you can bypass it when you know exactly what you want.

What it is

Jean Memory is a persistent memory layer for AI applications, reachable over MCP and through React, Python and Node SDKs. Its architecture has two levels: an orchestration tool that analyses the message and conversation history to decide a context strategy, and the granular tools underneath that it calls to do the work — which you can also call directly.

What you get
  • An intelligent entry point that analyses intent, gathers the right context, synthesises it into one package for the model, and handles background memory saving — `jean_memory`
  • The granular tools beneath it, exposed for when you want direct control — `add_memories`, `search_memory`, `list_memories`, `deep_memory_query`
  • The same tools in both local and production environments, so what you develop against is what you deploy against
  • SDKs that put a context-aware chat component into a React app, or fetch context in a Python or Node backend before you call your own LLM
  • A self-hostable stack — API, UI and databases — if you do not want the managed service
Requirements

For the managed service: an API key, and for MCP a user ID — clients connect through `supergateway` to `https://api.jeanmemory.com/mcp/claude/sse/YOUR_USER_ID_HERE`, with VS Code additionally needing `chat.mcp.enabled` and `chat.mcp.discovery.enabled` in settings. Running the whole stack locally needs Node.js 18+, Python 3.12+, Docker with Compose and Git; `make setup` collects the keys — `OPENAI_API_KEY` is required, `GEMINI_API_KEY` optional — then `make build` and `make dev` bring up the UI on port 3000 and the API on port 8765. The project incorporates code from mem0 under Apache 2.0; additions by Jean Technologies are proprietary.