Databricks Agent Bricks — Official MCP Server

This section covers how to deploy the Official Neo4j MCP Server using Databricks Apps.

By following this section, you can expose Neo4j based MCP Tools and interact with them using the Databricks Agent capabilities, enabling integration with LLM agents or other workflows.

A detailed Step-by-step guide covers in details all the aspects of the Databricks App implementation. The following guide lets you deploy the MCP server in minutes using automated scripts.

Quick Start

Prerequisites: Databricks CLI installed and authenticated, a Neo4j instance.

# 1. Authenticate with Databricks
databricks auth login --host https://<your-databricks-workspace>

# 2. Configure Neo4j credentials
cp .env.sample .env
# Edit .env with your Neo4j URI, username, and password

# 3. Upload secrets to Databricks
./setup_secrets.sh [--profile <databricks-profile>]

# 4. Deploy the app
./deploy.py --app-name mcp-neo4j [--profile <databricks-profile>]

The app name must start with mcp- for Databricks to treat it as an MCP server.

After deploying, open the Databricks Playground, add your MCP server from Tools > Add Tool > MCP Servers, and start querying.

To sync file changes without a full redeploy:

./deploy.py --app-name mcp-neo4j --sync [--profile <databricks-profile>]

Introduction

This guide demonstrates how to deploy the Official Neo4j MCP Server using Databricks Apps.

The setup allows you to use the official Neo4j MCP Tools to interact with a remote Neo4j instance directly from Databricks. By exposing Neo4j-based MCP Tools, you can integrate with LLM agents, the Databricks Playground, or other workflows.

Architecture Overview

Databricks Agent / Playground
  -> Databricks App (Official MCP Server)
    -> Proxy forwarding requests to neo4j-mcp-server package
      -> Neo4j Database (e.g., demo.neo4jlabs.com / companies dataset)

Key points:

  • The official neo4j-mcp-server Python package exposes tools to interact with Neo4j.

  • A proxy is the entry point for the Databricks App, controlling requests to the MCP Server.

  • The Neo4j connection is secured using SSL.

  • Credentials are stored as Databricks secrets.

Advantages:

  • No code / low infrastructure (Databricks App)

  • Fast prototyping with local testing

  • Automatic permission inheritance

  • Schema-level exposure (multiple functions as multiple tools)

  • Works in Playground immediately

Limitations:

  • Python only

Implementation

Step 1 - Configure Secrets

Copy the sample env file and fill in your Neo4j credentials:

cp .env.sample .env

The .env file requires:

NEO4J_URI=neo4j+s://<your-neo4j>:7687
NEO4J_USERNAME=
NEO4J_PASSWORD=

Upload the secrets to Databricks:

./setup_secrets.sh [--profile <databricks-profile>]

Step 2 - The MCP Server App

The app structure:

app/
  app.py                      # Uvicorn entry point / proxy
  app.yaml                    # Maps Databricks secrets to env vars
  requirements.txt            # Python dependencies
  neo4j_mcp_server_process.py # Launches and manages the MCP server process

  • app.yaml binds Databricks Secrets to environment variables.

  • neo4j_mcp_server_process.py launches the official MCP server as a subprocess.

  • app.py is a uvicorn app that proxies requests to the MCP server process, adding authentication headers.

You can test the server locally using the provided client:

python client.py

Step 3 - Deploy

Deploy using the deploy script, which uses Databricks Asset Bundles to create the app and bind secrets automatically:

./deploy.py --app-name mcp-<app_name> [--profile <databricks-profile>]

Check your Workspace to review the app and synced files. The App is associated with a Service Principal — ensure it has grants to read secrets.

Test and Use

Playground

In the Playground, select your MCP Server from Tools > Add Tool > MCP Servers. Add a system prompt such as:

Purpose: Assist users in getting companies/organizations info.

Limitations:
- Focus on companies.
- Be conversational but do not answer unrelated queries.
- Handle queries for multiple companies.
- If there is no company information, inform the user.

Data Sources:
- Use the mcp tools you have been provided when requested with questions about companies.

Sample Questions:
- "What are the competitors of 'BigFix'?"
- "Show me the top 3 software companies by revenue"

If the model says it cannot use the MCP Server, try switching to another model such as Claude.

External Use

Find the public URL for your app under Compute > Apps in your Databricks workspace.

To integrate the app externally, obtain a Databricks token:

databricks auth token -p <your-profile>

See client_workspace.py for a Python client example using Workspace authentication.

You can also publish your App to the Databricks Marketplace.

Other Guides