The year of agent in 2025 does not lack of new buzzwords and one of the latest trending three-character word is called MCP (and A2A). Unlike Agents, everything is literally called an agent in 2025, MCP is a very much needed standard in today’s Data and AI world. And it will further transform the Internet with powered agents!
What is MCP?
Once upon a time, Microsoft was the only cool company in the world and Bill Gates was still the CEO of Microsoft. Microsoft Certified Professional (MCP) is a highly sought after title. Fast-forward in 2025, MCP means Model Context Protocol.
MCP meaning shifted in the past 20 years
Google search trend illustrated the interest of both MCP in the past 20 years. MCP interest shot up because a series of announcements from big tech rallying to support MCP in their AI offering.
According to Wikipedia, "The Model Context Protocol (MCP) is an open standard, open-source framework introduced by Anthropic in November 2024 to standardize the way artificial intelligence (AI) systems like large language models (LLMs) integrate and share data with external tools, systems, and data sources.[1] MCP provides a universal interface for reading files, executing functions, and handling contextual prompts.[2]"
Simply saying, MCP is similar to REST API standard for API tools, which is pretty much the way to common recognized way to communicate on the internet. MCP implements JSON-RPC 2.0 to allow the execution of the tools for AI models.
MCP architecture
Similar to REST API, MCP processes are hosted on a server. However, as opposed to REST API, which is usually being used by an application to perform specific functions, MCPs are being leveraged by AI models for tool calling, aka Agentic workflow.
MCP architecture (source: Wikipedia)
Databricks as an MCP Server
Databricks provides out of the box MCP Servers integrated with Unity Catalog. These are called Managed MCP Servers. On the other hand, as discussed before, Databricks Apps are no more than a web server with tight security control. Naturally, we can host MCP Servers on Databricks Apps.
There are three types of Managed MCP Servers:
- Vector search
It allows agents query Databricks Vector Search indexes in the specified Unity Catalog schema.
url:https://<your-workspace-hostname>/api/2.0/mcp/vector-search/{catalog_name}/{schema_name} - Unity Catalog functions
It allows agents to run Unity Catalog functions in the specified Unity Catalog schema.
url:https://<your-workspace-hostname>/api/2.0/mcp/functions/{catalog_name}/{schema_name} - Genie space
It allows agents to query the specified Genie space to get insights from structured data (tables in Unity Catalog)
url:https://<your-workspace-hostname>/api/2.0/mcp/genie/{genie_space_id}
Let’s put it in action
Enough theories. It’s time to put everything in action. We will leverage the MCP inspector to inspect Databricks UC functions in MCP.
MCP Inspector
It’s relatively easy to setup and test the MCP endpoints.
- Transport Type: Streamable HTTP (more info here: https://mcp-framework.com/docs/Transports/transports-overview/)
- URL: https://adb-xxxx.azuredatabricks.net/api/2.0/mcp/functions/system/ai
- Authentication Header Name: Authorization
- Authentication Bearer Token: Your Databricks PAT
- Proxy Session Token: As shown in the console
The above screenshot illustrates the settings. And in this example, we wanted to test our the system.ai functions within Databricks. This is the scenario #2 in the section above "Databricks as an MCP Server".
Finally, after successfully connection, we can click on the "List Tools" button and try out some of the tools. Below is an example of a successful execution.
MCP tool execution
Conclusion
MCP is an evolving standard that a lot of Enterprises like OpenAI, Google and Microsoft have adopted. It’s similar to REST API but really the other way around that allows AI tools to execute the functions or tools hosted online. Databricks has opened up a way by bringing in Managed MCP Servers tightly integrated with Unity Catalog. There is no setup or coding required. If developers wanted to host their own MCP Server, Databricks Apps is there to serve this purpose.
By testing MCP Inspector (developed by Anthropic) and being able to connect to Databricks, it illustrates that Databricks’ commitment to open standards.
If you are ready to see more real life example, check out the below OpenAI cookbook.
https://cookbook.openai.com/examples/mcp/databricks_mcp_cookbook
AUTHOR - FOLLOW
Jason Yip
Director of Data and AI, Tredence Inc.
