neo4j-field/mcp-neo4j-vectordb

Neo4j VectorDB MCP Server

An MCP server that uses Neo4j as a pure vector database — no graph traversal, no relationships.

Purpose

This server is designed to compare Vector RAG vs Graph RAG. It provides the same search capabilities as mcp-neo4j-graphrag but intentionally hides graph features, allowing you to measure the value that graph context adds to LLM responses.

Installation

Step 1: Download the Repository

git clone https://github.com/guerinjeanmarc/mcp-neo4j-vectordb.git
cd mcp-neo4j-vectordb

Step 2: Configure Claude Desktop

Edit the configuration file:

• macOS/Linux: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this server configuration (update the path to where you cloned the repo):

{
  "mcpServers": {
    "neo4j-vectordb": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-neo4j-vectordb",
        "run",
        "mcp-neo4j-vectordb"
      ],
      "env": {
        "NEO4J_URI": "neo4j+s://demo.neo4jlabs.com",
        "NEO4J_USERNAME": "recommendations",
        "NEO4J_PASSWORD": "recommendations",
        "NEO4J_DATABASE": "recommendations",
        "OPENAI_API_KEY": "sk-...",
        "EMBEDDING_MODEL": "text-embedding-ada-002"
      }
    }
  }
}

Step 3: Reload Configuration

Quit and restart Claude Desktop to load the new configuration.

Tools

get_searchable_content

Discover available indexes and searchable node properties (no relationships shown).

💡 The agent should automatically call this tool first to understand what can be searched.

Example prompt:

"What can I search in this database?"

vector_search

Semantic similarity search.

Example prompt:

"Find movies about 'a hero's journey in space'"

fulltext_search

Keyword search with Lucene syntax.

Example prompt:

"Where was Tom Hanks born??"

get_node_by_id

Retrieve a node's properties by ID (no relationships).

Example prompt:

"Get the full details of the first movie from my search results"

Comparison Workflow: Vector RAG vs Graph RAG

Use both servers to compare how graph context improves LLM responses.

Prerequisites

Download both repositories:

git clone https://github.com/guerinjeanmarc/mcp-neo4j-vectordb.git
git clone https://github.com/guerinjeanmarc/mcp-neo4j-graphrag.git

Step 1: Configure Both Servers in Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "neo4j-vectordb": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-neo4j-vectordb",
        "run",
        "mcp-neo4j-vectordb"
      ],
      "env": {
        "NEO4J_URI": "neo4j+s://demo.neo4jlabs.com",
        "NEO4J_USERNAME": "recommendations",
        "NEO4J_PASSWORD": "recommendations",
        "NEO4J_DATABASE": "recommendations",
        "OPENAI_API_KEY": "sk-...",
        "EMBEDDING_MODEL": "text-embedding-ada-002"
      }
    },
    "neo4j-graphrag": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-neo4j-graphrag",
        "run",
        "mcp-neo4j-graphrag"
      ],
      "env": {
        "NEO4J_URI": "neo4j+s://demo.neo4jlabs.com",
        "NEO4J_USERNAME": "recommendations",
        "NEO4J_PASSWORD": "recommendations",
        "NEO4J_DATABASE": "recommendations",
        "OPENAI_API_KEY": "sk-...",
        "EMBEDDING_MODEL": "text-embedding-ada-002"
      }
    }
  }
}

Step 2: Test with VectorDB Only

1. In Claude Desktop, click the "Search and Tools" button in the conversation interface
2. Disable neo4j-graphrag tools (keep only neo4j-vectordb active)
3. Ask the test question:

"Find movies about artificial intelligence, and tell me which directors have made multiple AI films and what other genres they typically work in."

4. Note Claude's answer and what tools it used

Step 3: Test with GraphRAG Only

1. Click "Search and Tools" again
2. Disable neo4j-vectordb tools and enable neo4j-graphrag tools
3. Ask:

"You have now access to the Neo4j tools, please answer the same question."

4. Note Claude's answer and what tools it used

Step 4: Compare Results

Ask Claude to compare:

"Compare your two previous answers. What approach gave you better context to answer the question? Please explain what technology is better to answer the question: VectorDB or Neo4j?"

Example Questions for Comparison

Configuration

Environment Variables

Embedding Providers

Supports all LiteLLM embedding providers:

• OpenAI: text-embedding-ada-002, text-embedding-3-small
• Azure: azure/deployment-name
• Bedrock: bedrock/amazon.titan-embed-text-v1
• Cohere: cohere/embed-english-v3.0
• Ollama: ollama/nomic-embed-text

Related

• mcp-neo4j-graphrag — Full GraphRAG with graph traversal
• Neo4j MCP Documentation
• Official Neo4j MCP Server

License

MIT License