Semantic Search MCP Server

Universal, schema-agnostic semantic search for any Neo4j knowledge graph.

One tool. Works everywhere. No configuration needed.

What It Does

Searches across ALL vector indexes in your Neo4j database using semantic similarity. Returns anything that matches your natural language query above a threshold.

Works with:

• Any node type
• Any relationship type
• Any schema
• Any knowledge graph with embeddings

Installation

1. Install Dependencies

pip install -r mcp_servers/semantic_search/requirements.txt

2. Configure Environment Variables

# Required
export OPENAI_API_KEY="sk-..."
export NEO4J_PASSWORD="your-password"

# Optional (defaults shown)
export NEO4J_URI="bolt://localhost:7687"
export NEO4J_USER="neo4j"
export EMBEDDING_MODEL="text-embedding-3-small"

3. Register with Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "semantic-search": {
      "command": "python",
      "args": ["-m", "mcp_servers.semantic_search.server"],
      "cwd": "/path/to/your/e2r-rd",
      "env": {
        "OPENAI_API_KEY": "sk-...",
        "NEO4J_PASSWORD": "your-password"
      }
    }
  }
}

4. Restart Claude Desktop

Usage

The Tool: semantic_search

One simple tool with three parameters:

semantic_search(
    query: str,          # Natural language query
    limit: int = 10,     # Max results (max: 100)
    threshold: float = 0.7  # Min similarity (0-1)
)

Examples:

semantic_search("database failures")
semantic_search("kubernetes scaling", limit=20)
semantic_search("cache problems", threshold=0.8)

Returns:

Found 5 results matching 'database failures':

1. **Redis Cache** [Service] (similarity: 0.923) - mentioned 15 times
2. **PostgreSQL** [Service] (similarity: 0.887) - mentioned 12 times
3. **cache timeout** [Action] (similarity: 0.845) - mentioned 8 times
4. **database connection pool** [GenericEntity] (similarity: 0.821)
5. **connection failures** [Metric] (similarity: 0.798) - mentioned 6 times

How It Works

1. Auto-discovers all vector indexes in your database
2. Generates one embedding for your query (via OpenAI)
3. Searches all indexes in parallel
4. Filters by threshold
5. Returns top results sorted by similarity

No schema knowledge required!

Prerequisites

Your Neo4j database needs:

1. Nodes with embeddings - Any nodes with an embedding property (list of floats)
2. Vector indexes - Created on the embedding properties

Create Vector Indexes

CREATE VECTOR INDEX my_index_name IF NOT EXISTS
FOR (n:MyNodeLabel) ON (n.embedding)
OPTIONS {indexConfig: {
    `vector.dimensions`: 1536,
    `vector.similarity_function`: 'cosine'
}}

Repeat for each node type that has embeddings.

Testing

# Set environment variables
export OPENAI_API_KEY="sk-..."
export NEO4J_PASSWORD="your-password"

# Run test script
python mcp_servers/semantic_search/test_search.py

Troubleshooting

No results found

Check:

1. Neo4j is running
2. Vector indexes are created (SHOW INDEXES in Neo4j)
3. Nodes have embeddings populated
4. Try lowering the threshold (default: 0.7)

Connection errors

Check:

• Neo4j URI is correct
• Neo4j password is correct
• Network connectivity
• Neo4j is running: docker ps or Neo4j Desktop

MCP server not appearing

Check:

1. Claude Desktop config JSON is valid
2. cwd path is absolute and correct
3. Claude Desktop was restarted
4. Check Claude Desktop logs for errors

Architecture

semantic_search/
├── config.py            # Environment configuration
├── semantic_search.py   # Core search engine
├── server.py            # MCP server
├── test_search.py       # Test script
└── README.md            # This file

Key Features

• Schema-agnostic: Discovers indexes at runtime
• Single embedding: Generated once, used everywhere
• Parallel search: Searches all indexes simultaneously
• Threshold filtering: Only returns good matches
• Error resilient: Continues if one index fails
• Simple interface: One tool, three parameters

Configuration Reference

Environment Variables

# Required
OPENAI_API_KEY         # OpenAI API key for embeddings
NEO4J_PASSWORD         # Neo4j password

# Optional
NEO4J_URI              # Default: bolt://localhost:7687
NEO4J_USER             # Default: neo4j
EMBEDDING_MODEL        # Default: text-embedding-3-small
DEFAULT_LIMIT          # Default: 10
MAX_LIMIT              # Default: 100

Tool Parameters

Parameter	Type	Default	Description
query	string	required	Natural language query
limit	integer	10	Max results (max: 100)
threshold	float	0.7	Min similarity (0-1)

Threshold Guide

• 0.9+: Very strict, near-exact matches
• 0.8-0.9: Strong semantic similarity
• 0.7-0.8: Good matches (default range)
• 0.6-0.7: Loose matches, more results
• <0.6: Very loose, may include noise

Examples

Basic Search

semantic_search("monitoring alerts")

Strict Matching

semantic_search("postgresql database", threshold=0.85)

Broad Exploration

semantic_search("cloud infrastructure", limit=50, threshold=0.6)

Development

To modify the tool:

1. Edit semantic_search.py to change search logic
2. Edit server.py to change tool interface
3. Run test_search.py to validate

License

Part of the e2r-rd project.