nocyphr/OpenMem0MCP

OpenMem0MCP

OpenMem0MCP is a fast, pluggable memory control server that wraps the mem0 Python package and exposes a simple HTTP interface for long-term memory operations: memorize (add) and remember (search).

This project is designed to be modular, production-ready, and highly configurable, enabling you to swap out the memory backend, vector store, embedding provider, and graph store — all through environment variables.

This package includes two companion patches that make the mem0 Python SDK suitable for real-world, concurrent, production environments.

What This Is

• A Memory Control Protocol (MCP) server exposing /add and /search endpoints.
• Wraps mem0 with extended backend support.
• Includes containerized support for:
  • PostgreSQL for relational memory storage.
  • Milvus for high-performance vector similarity search.
  • Neo4j for optional graph-based memory linkage.
• Supports concurrent writes, unlike SQLite.
• Flexible embedding model support beyond what mem0 supports natively.

Key Enhancements Over mem0

Patch #1 – SQLAlchemy Backend Expansion

mem0_history_db_patch

• Extends mem0 to support any SQLAlchemy-compatible database.
• Enables PostgreSQL as a robust alternative to SQLite for concurrent environments.
• Included via a pip-installable companion patch.

Patch #2 – Expanded Embedding Provider Support

mem0_embeddings_litellm_patch

• Integrates litellm as an embedding backend for mem0.
• Enables support for a wider range of embedding providers and models.
• Offers an easy switch between providers like OpenAI, HuggingFace, and others.

Together, these patches make the original mem0 project suitable for production usage in multi-agent or high-load environments.

Setup Instructions

1. Clone & Configure

git clone https://github.com/nocyphr/OpenMem0MCP.git

Edit .env and provide the necessary values:

2. Run with Docker Compose

docker-compose -f docker/docker-compose.yml up --build

This launches:

• The MCP server
• Postgres
• Milvus
• Neo4j (optional)

3. Test the Endpoints

Use the provided testme.py script to verify your setup:

python3 -m venv .venv
source .venv/bin/activate
pip install -r docker/requirements.txt
python testme.py

Alternatively in src/client.py is a sample script that makes an agent use the tools. You can run that too for an agentic test. This sends test requests to your running MCP server.

API Overview

Note: Deletion and manual updates are not exposed as endpoints. mem0 handles internal consistency and updates automatically when new, conflicting data is added.

Modular Architecture

You can configure the system to use:

• Any SQLAlchemy-supported database (Supabase/PostgreSQL, MySQL, etc.)
• Any embedding model supported by litellm
• Any Graphstore supported by mem0 (default setup is neo4j)
• Any Vectorstore supported by mem0 (default setup is milvus)

All components are swappable through .env configuration.

Usage Note

At this time, the project is designed to run out-of-the-box with:

• MilvusDB (as vector store)

• PostgreSQL (as history database)

• Neo4j (as Graphstore) If you want to use different databases, graphstores or vector stores, you'll need to either:

• Provide access to an already-hosted instance via environment variable URLs, or

• Add the appropriate container and configuration to the existing docker-compose.yml.

In the future, additional compose files may be provided for different deployment configurations, but for now, a certain level of manual setup is assumed if you don't want to use the readymade config.

Roadmap / To-Do

• Finalize .env migration for all config variables
• remove ports mapping to host for companion services
• Add Traefik reverse proxy
• add option for raw ingestion to vstore
• add tests (duh)

Credits

Created and maintained by [Nocyphr]. Built on top of the excellent mem0 project, extended and productionized via custom patches and modular deployment design.