Databricks Integration Examples

A collection of practical integration patterns for developers and ISV partners building on the Databricks platform. Each example focuses on a specific Databricks capability and demonstrates how to connect to it, query it, or extend it from outside the platform.

Examples

Agents — MCP Integration with OAuth

Connect an external AI agent to a Databricks Managed MCP server using OAuth M2M authentication. Demonstrates tool discovery and remote execution via the Model Context Protocol — the foundation for wiring external agents (LangChain, LangGraph, etc.) into Databricks capabilities.

---

AI/BI — Genie Conversations & Dashboard Metadata

Two patterns for programmatic interaction with Databricks AI/BI:

• Chat with a Genie Space via the Conversation REST API — embed natural language data querying into any application
• Extract and audit dashboard metadata across a workspace using the Databricks SDK

---

Model Serving — Calling Databricks AI Endpoints

Call Databricks Mosaic AI Model Serving endpoints using three patterns: OpenAI Python SDK, Databricks SDK, and REST API. Includes DBRX Pay-Per-Token inference and MLflow model metadata extraction. The go-to reference for integrating LLM or agent endpoints into external applications.

---

Unity Catalog — Lineage & Metadata

Two Unity Catalog integration patterns:

• Bring Your Own Lineage (BYOL): Register external systems (Salesforce, Qlik, Fivetran, etc.) as lineage participants in UC to build end-to-end data lineage visible in Catalog Explorer
• PK/FK Constraints: Define and query primary key / foreign key constraints on Delta tables for data modeling and BI integrations

---

Vector Search — Semantic Search with Databricks

Full lifecycle example for Databricks Vector Search using the Python SDK — create an endpoint, build a Delta Sync index, trigger syncs, and run similarity queries. The foundation for building RAG pipelines or semantic search over enterprise data in Unity Catalog.

---

Prerequisites

All examples run as Databricks notebooks. You will need:

• A Databricks workspace (AWS, Azure, or GCP)
• Unity Catalog enabled (for Model Serving, Vector Search, and UC examples)
• A personal access token or service principal credentials
• Feature-specific prerequisites are listed in each subfolder's README

Repository Structure

src/
├── agents/           # MCP + OAuth M2M integration
├── ai_bi/            # Genie Conversation API + dashboard metadata
├── model_serving/    # Model Serving endpoint patterns
├── unity_catalog/    # BYOL lineage + PK/FK constraints
└── vectorsearch/     # Vector Search SDK lifecycle