nargesghv/Insert-data-in-to-Neo4j-mcp-server

📘 Inserting Nodes and realations in to Neo4j(Graph Database) by applying MCP Server

A project that accepts natural language (NL) queries from Claude Desktop via MCP and uses GPT-4o + LangChain to extract structured node information and store it in a Neo4j database.

🚀 Features

• Neo4j graph database: stores nodes extracted from NL input.
• MCP server (FastMCP): offers an LLM-exposable tool createNode.
• LangChain + GPT‑4o: parses NL into structured Neo4jNode.
• Claude Desktop integration: uses MCP protocol to call your tool interactively.

🛠️ Prerequisites

• Python 3.10+
• Neo4j Community or Enterprise Edition (locally or via Aura)
• Claude Desktop installed
• uv CLI (optional but helpful for running MCP apps)

🗄️ Step 1: Install & Set Up Neo4j

1. Download and install Neo4j Desktop from neo4j.com.

2. Create a new database using the Neo4j Desktop UI or Neo4j Browser:

   CREATE DATABASE mydb IF NOT EXISTS;
   

3. Set credentials during creation (neo4j username and password).

4. Get the Bolt connection URI, typically:

   bolt://localhost:7687
   

🧩 Step 2: Project Configuration

1. Clone or create this project directory.

2. Create a .env file in the project root:

   uri=bolt://localhost:7687
   username=neo4j
   pwd=your_password
   

3. Install dependencies:

   pip install neo4j neo4j_graphrag langchain langchain-core langchain-openai mcp python-dotenv
   

🧱 Step 3: Code Overview

The server.py file includes:

• Prompt definition using LangChain's PromptTemplate and PydanticOutputParser
• Tool chain with GPT‑4o to convert NL to structured node
• FastMCP server exposing the createNode tool
• Neo4jWriter to persist graph nodes

🧪 Step 4: Run the MCP Server

You can run the MCP server using Python directly:

fastmcp install server.py

Or optionally using uv:

uv add mcp
uv run python server.py

💬 Step 5: Use with Claude Desktop (Client)

1. Launch Claude Desktop.

2. Connect it to your running MCP server (via standard I/O or HTTP).

3. Send a message like:

   Create a person node with name Alice and age 30
   

4. The tool will:

   • Generate a Neo4jNode from GPT‑4o

   • Write it to the Neo4j database

   • Respond with:

     ✅ Node successfully added to db — Label: Person, Properties: name:Alice, age:30
     

🔍 Step 6: Inspect Data in Neo4j

You can explore the created nodes using Neo4j Browser or Neo4j Bloom:

MATCH (n) RETURN n;

🔧 Step 7: Customize & Extend

• Add more tools (e.g., createRelationship, queryGraph)
• Integrate a FastAPI or Gradio UI for user interaction
• Use Claude with structured function calling via MCP

📚 References

• Neo4j Quickstart Docs
• Neo4j Cypher Manual
• FastMCP GitHub
• LangGraph MCP Protocol Guide
• Model Context Protocol

✅ Summary

This project brings together:

• 💬 Natural Language → Graph with GPT‑4o
• 📡 MCP tool exposed to Claude Desktop
• 🌐 Neo4j for persistent graph storage
• 🔁 Fully extendable graph-based RAG architecture

For any issues or ideas, feel free to open an issue or PR!