Text2Cypher ReAct Agent Example

This repo provides a simple example of how to build a ReAct agent with MCP and local tools.

Specifically this repo builds a Text2Cypher ReAct agent with the Neo4j Cypher MCP Server. The agent also has extended capabilities by exposing local tools for it to use.

A conversational AI agent that connects to a Neo4j Movies database and can answer movie-related questions using natural language. Built with LangGraph's implementation of a ReAct agent, the Neo4j Cypher MCP server, and a custom movie recommendations tool.

Features

• Natural Language to Cypher: Ask questions in plain English and get answers from your Neo4j database
• ReAct Agent Pattern: Uses reasoning and acting loops for complex reasoning
• Schema-Aware: Automatically retrieves and uses database schema for accurate query generation
• Interactive CLI: Chat-based interface for easy interaction

Prerequisites

• Python 3.10 or higher
• Neo4j Aura account or local Neo4j instance with Movies database
• OpenAI API key
• uv package manager (recommended) or pip

Installation

Option 1: Using uv (Recommended)

1. Install uv (if not already installed):

   Install Documentation

   pip install uv

2. Clone and setup the project:

   git clone neo4j-field/text2cypher-react-agent-example
   cd text2cypher-react-agent-example

3. Install dependencies:

   uv sync

Option 2: Using pip

1. Clone and setup the project:

   git clone neo4j-field/text2cypher-react-agent-example
   cd text2cypher-react-agent-example

2. Create and activate virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

Configuration

1. Copy the example environment file:

   cp .env.example .env

2. Edit .env with your credentials:

   OPENAI_API_KEY=your_openai_api_key_here
   NEO4J_USERNAME=neo4j
   NEO4J_PASSWORD=your_neo4j_password
   NEO4J_URI=neo4j+s://your-instance.databases.neo4j.io
   NEO4J_DATABASE=neo4j

Agent Components

Core Agent (agent.py)

• LangGraph ReAct Agent: Implements reasoning and acting loops for complex queries
• Neo4j Cypher MCP Server: Provides schema introspection and query execution
• Custom Recommendations Tool: Custom tool for movie recs
• Interactive CLI: Command-line chat interface

Key Tools Available to the Agent

1. get_neo4j_schema: Retrieves database schema for informed query writing
2. read_neo4j_cypher: Executes read-only Cypher queries against the database
3. find_movie_recommendations: Custom recommendation engine that finds movies liked by users who also enjoyed a target movie

Running the Agent

Using uv (Recommended)

make run-agent-uv
# or
uv run python3 single_file_agent.py

Using pip/standard Python

make run-agent
# or
python3 single_file_agent.py

Usage Examples

Once running, you can ask questions like:

• "What movies are in the database?"
• "Tell me about The Matrix"
• "Recommend me some films like The Dark Knight."

Exit Commands

To exit the agent, type any of:

• exit
• quit
• q

Evaluation

This repo also contains a simple local evaluation suite. This may be used and extended to evaluate your own agents.

Running Evaluations

To run

1. Configure the eval.py file with the LLM name, tools and prompt you would like to use
2. Ensure you have populated the questions.yaml file with your eval question set
3. Run make run-eval-uv or make run-eval depending on your package manager
4. The eval results CSV will be saved to evals/output/<file-name>.csv
5. View the contents with review.ipynb
6. Generate a .txt report with make generate-report-uv csv-name=<file-name> or make generate-report csv-name=<file-name> depending on your package manager

Eval CSV Structure

The resulting evaluation CSV will contain the following columns:

• question_id: str
• question: str
• expected_answer: str
• agent_final_answer: Optional[str]
• generated_cypher: list[ReadNeo4jCypherToolInput]
• model: str
• available_tools: list[str]
• called_tools: list[str]
• num_messages: Optional[int]
• num_llm_calls: Optional[int]
• num_tool_calls: Optional[int]
• response_time: Optional[float]
• error: Optional[str]

Development

Code Formatting

make format

Dependencies

Core Libraries:

• langchain - LangChain framework
• langchain-mcp-adapters - MCP (Model Context Protocol) adapters
• langchain-openai - OpenAI integration
• langgraph - Graph-based agent framework
• neo4j - Neo4j Python driver
• openai - OpenAI API client
• pydantic - Data validation

Development:

• ruff - Code formatting and linting

Troubleshooting

Connection Issues:

• Verify your Neo4j credentials in .env
• Ensure your Neo4j instance is running and accessible

OpenAI Issues:

• Verify your OpenAI API key is valid
• Check your API usage limits

MCP Server Issues:

• Ensure uvx is available in your PATH
• The agent automatically installs [email protected] via uvx

Python Issues:

• Ensure Python 3.10+ is installed
• Try recreating your virtual environment if using pip

License

This project is provided as an example for educational and demonstration purposes.