🧠 Memora MCP Server

Memora MCP Server is a lightweight backend built using FastAPI and Neo4j that acts as a memory engine — capable of storing entities and relationships representing personal interests, skills, and experiences as a Knowledge Graph. This project demonstrates how MCP (Model Context Protocol) can be used to give LLMs persistent memory and reasoning over structured context.

---

🚀 Features

• 🧩 Entity and relationship creation with Neo4j
• ⚡ REST API built with FastAPI
• 🤖 Integration-ready MCP endpoint for LLM clients (like Claude)
• 🧠 Designed as a base for building personalized AI memory or contextual intelligence systems

---

⚙️ Setup

1. Clone the repository

git clone https://github.com/Deepak3168/Memora.git
cd memora-mcp-server

2. Create a virtual environment

Use venv to keep dependencies simple and isolated:

python -m venv mcp
source mcp/bin/activate   # (Linux/Mac)
mcp\Scripts\activate      # (Windows)

3. Install dependencies

pip install -r requirements.txt

4. Run the FastAPI server

uvicorn main:app --host 0.0.0.0 --port 8000 --reload

Your API will now be available at: 👉 http://127.0.0.1:8000/docs

---

🗄️ Neo4j Setup

1. Install Neo4j

sudo apt install neo4j -y

2. Change Neo4j password

Use this terminal command:

cypher-shell -u neo4j -p neo4j "ALTER CURRENT USER SET PASSWORD FROM 'neo4j' TO 'password';"

3. Start and enable service

sudo systemctl start neo4j
sudo systemctl enable neo4j

4. Create .env file

Create a .env file as mentioned in constants.py:

NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_password_here

---

🧩 MCP Server Integration (Claude)

You can connect this backend to Claude Desktop as an MCP Server.

Steps:

1. Open Claude → Settings → Developers → Edit Config

2. Copy the JSON config from:

   claude_desktop_config.json
   

3. Restart Claude

   • On Windows: close it from Task Manager
   • On Mac/Linux: simply restart the app

You should now see Memora MCP Server listed in Claude’s connected MCP tools.

---

💬 Example Prompt for Claude

Try this inside Claude after connecting the MCP server:

“From now, create entities and relationships based on my interests, skills, and knowledge. Build a knowledge graph that reflects my personality, expertise, and learning history.”

Claude will then begin creating entities (representing your interests, skills, etc.) and relationships (showing how they connect), gradually forming a personal knowledge graph.

---

🔮 Potential Use Cases

• 🧠 Personalized AI assistants that remember your context
• 🎯 Context-driven recommendations and ads
• 🔍 Graph-to-vector embeddings for LLM fine-tuning
• 🧩 Building a long-term AI memory system

---

🧱 Project Structure

📦 Memora
├── main.py                      # FastAPI entry point
├── routes/
│   ├── entity_ops.py            # Entity creation & listing routes
│   └── relation_ops.py          # Relationship endpoints
├── models/
│   └── entity.py           # EntityNode definition
|   └── relation.py           # RelationEdge definition
├── db/
│   └── init.py                  # Neo4j connection setup
├── constants.py                 # Env and configuration constants
├── mcp_config.json   # MCP config for Claude
└── requirements.txt

---

🧠 Concept Overview

Every time you interact with the MCP server, new entities (like Deepak, Backend Developer, Python) and relationships (like Deepak → skilled_in → Python) are stored in Neo4j.

Over time, this forms a graph representation of your memory and interests — a foundational step toward a Memory Engine that can later:

• Generate embeddings
• Train fine-tuned models
• Enable contextual recall for future prompts

---

🧑‍💻 Author

Deepak Ootala Passionate about backend development, data engineering, and building intelligent systems that evolve with user context.