New developments at Databricks: How managed MCP servers make AI tool integration secure and efficient

The dynamic development of AI models, especially Large Language Models (LLMs), is increasingly opening up opportunities for companies to integrate complex automated agents that can actively support their business processes. In order to be able to represent the potential of these models, a new standard has recently been established - the Model Context Protocol (MCP).

At the same time, however, many companies are facing major challenges: How can these technologies be introduced without compromising data security, governance and discoverability? This is exactly where Databricks' latest innovation comes in - the launch of Managed MCP Servers in conjunction with Unity Catalog and Mosaic AI. This new combination enables the enterprise sector to use MCP smoothly and, above all, securely.

What exactly is the Model Context Protocol (MCP)?

Before we delve deeper into the innovations of Databricks, let's briefly clarify what MCP actually is and why it is currently receiving attention across the industry. MCP is establishing itself as a comprehensive standard that makes it possible to equip large language models and AI agents with additional tools and contextual knowledge. Previously, LLMs were mostly isolated from corporate data; LLMs can now directly access context-specific data sources and APIs through MCP and thus carry out natural language-based actions.

But MCP in itself does not solve all challenges. Companies are confronted with questions about authentication, authorization and data protection, particularly during integration. Databricks is addressing precisely these challenges with its new approach, which combines MCP integration with existing Databricks technologies.

How Databricks Managed MCP Servers solve market challenges

With the introduction of Managed MCP Servers, Databricks connects the MCP concept closely with the already established services Unity Catalog and Mosaic AI:

• Unity Catalog for Governance and Discovery: This component ensures that all data is centralized and organized under strict compliance and privacy policies.
• Mosaic AI for Agent Development and Evaluation: Mosaic AI provides a robust framework to create, evaluate, and continually improve AI agents.
• Managed MCP Servers for secure tool integration and access management: The new Managed MCP Servers automatically include the so-called “on-behalf-of-user” authentication model, which integrates seamlessly into existing Unity Catalog security mechanisms. This ensures that users only see the data for which they are authorized at any time - a real added value, especially for highly regulated industries.

In addition, Databricks' MCP offering also includes standard support for OAuth via Databricks Apps, making it quicker and easier to connect existing services and APIs to the MCP architecture.

Typical application example: A smart customer support agent in the telecommunications environment

Let's illustrate the potential of managed MCP servers using a concrete scenario: A telecommunications company wants to support its customer advisors with AI-based agents. These AI agents are intended to read internal data from the Unity Catalog, e.g. B. about past interactions or products, as well as current real-time information - for example reports of outages or network disruptions. Until now, this data was often “locked up” in company-internal platforms and APIs.

Thanks to Databricks Apps including standard OAuth, Git-based provisioning, and built-in governance, these existing APIs can now be deployed as MCP servers in no time. The AI ​​agent can then “interview” internal services directly to offer customers personalized and up-to-date information. Using Databricks' serverless infrastructure also ensures scalability without additional effort for IT teams.

Advantages and opportunities for companies thanks to Databricks Managed MCP Servers

The introduction of Managed MCP Servers by Databricks offers companies clear advantages and significant opportunities:

• Security & Compliance: Full integration with Unity Catalog allows organizations to tightly control access to sensitive data without having to build additional layers of governance.
• Faster Development & Time-to-Value: The pre-built templates and fast API integration dramatically reduce the effort for developers - they can focus more on business logic instead of solving complex network and authentication problems.
• Scalability without additional maintenance: The managed service means no additional maintenance or care for your IT department. Databricks takes care of updating and maintaining the MCP servers and ensures that the infrastructure always remains state-of-the-art.
• Flexibility & future security: A variety of services such as Genie, Vector Search and UC Functions can already be controlled today. Databricks also plans to continuously expand the MCP server functionality with new resources such as DBSQL and externally provided MCP servers from other providers.

Prototyping and monitoring via AI Playground and MLflow integration

Another highlight of this solution lies in its extensive prototyping and monitoring tools: The AI ​​Playground allows companies to quickly and easily develop mayor-approved agents and test them in a secure environment before a line of real code is written. Using an integrated MLflow evaluation, users simultaneously document how internal tools are used and gain a deeper understanding of agent operations in the company.

This allows you to quickly and continuously adapt your models based on real performance data. This in turn maximizes the efficiency and accuracy of AI systems in the production environment.

Conclusion: Databricks creates a secure and efficient future for AI and data integration

Databricks' Managed MCP Servers represent a milestone when it comes to using Large Language Models sustainably, securely and usefully in the enterprise environment. Enterprises gain the opportunity to leverage state-of-the-art AI integration without compromising governance and security.

The beta version of Managed MCP Servers is already available - companies should take the opportunity to familiarize themselves with this powerful extension early on in order to secure a clear competitive advantage in their respective industries.

For companies that want to focus more on data engineering, industrial AI and high-performance AI integration, this innovation from Databricks represents an essential further development in order to be even more successful and innovative in the future.