# Ontolocy MCP Server

Simple proof of concept MCP server to augment an LLM with Ontolocy cyber graph capabilities.

Key technologies:

* Neo4j - for the graph database
* MCP Python SDK - for the MCP server
* PydanticAI - for the client (which connects to the MCP server)
* Neontology - Interacting with Neo4j using pydantic models
* Ontolocy - Cyber graph ontology and tooling built with Neontology

## Quickstart

### Pre-Requisites

Install dependencies with `uv`:

```bash
uv sync
```

Make sure you have a .env file which specifies:

```.env
FASTMCP_PORT=8008

OPENAI_API_KEY=<YOUR API KEY>

NEO4J_URI="neo4j://localhost:7687"
NEO4J_USERNAME=neo4j
NEO4J_PASSWORD=<YOUR NEO4J PASSWORD>
```

Ensure your Neo4j database is running and accessible. Be warned, the LLM may run arbitrary queries against it!

### Populate the Graph

Use Ontolocy to populate the graph with MITRE ATT&CK data from the official github repo.

```bash
uv run ontolocy parse mitre-attack
```

### Start the Server

```bash
uv run server.py
```

This will make the MCP server available on the port configured in the .env file (e.g.8008).

### Use the CLI Client

```bash
uv run mcp_sse_client.py lookup mitre attack technique T1090 in the ontolocy cyber graph
```