Introduction

We recently introduced Model Context Protocol (MCP) in Collate, the details of which are covered in this blog. A visual run-through is available in this YouTube video. This blog is going to run you through some of those examples to show you what is possible and to get you thinking about additional ways you can get out of the box and create new and exciting interactions with Collate.

I’m not going to cover how to enable MCP in Collate in this blog, but this short video will show you how to do it, and it is in our docs as well. In addition, I will put a link to the corresponding video for each segment.

With that intro out of the way, let’s dive right into it.

What are some things we can do?

Creating Glossaries with AI

[Video] Let’s start with building a glossary. As you can see, I don’t have one yet, and creating one from scratch is a time-consuming process.

Using Claude, we simply ask “Can you create an ecommerce glossary”. Claude helpfully agrees to the task, and sees that we have no glossaries yet, and begins its work:

After about a minute, we get some results:

That’s all well and good, but what do we see in Collate?

The glossary terms have now been added. We can get more elaborate though, and include external sources to ingest as well, for example:

I ask Claude to add glossary terms from this external source to my eCommerce glossary that was just created. Now I see that I have 28 glossary terms:

What’s important to note is that this isn’t limited to what is internally known by Collate, or a URL. You can take advantage of Collate's bulk upload feature to upload a CSV file of terms, which will then be applied, thereby greatly reducing your repetitive or mundane workload.

Monitoring Pipelines with AI

[Video] Since we’ve already shown how the Collate MCP server can write data back to Collate, let’s focus on a few fun use cases and how those look. Imagine you're a data engineer, and you want to build a dashboard that shows all the pipelines and statuses in your system. Using the following prompt:

“Get me a status of all the pipelines, and prepare a table with the pipelines and corresponding status for them.”:

We get a really robust response, with a lot of useful information, which becomes exponentially more helpful the more you are dealing with. We can think outside the box a little bit, though, and ask Claude to “create a modern interface for the data”, and after just a few seconds, we see:

That’s pretty slick, and it’s easy to play around with the prompt to customize it further.

Additional ideas for AI prompts

[Video] Test cases can be a tedious task to put together, but MCP-powered Collate can make this simple. Consider the following prompt:

“Can you find the CustomerInfo table and add relevant test cases at the table and column level?”

Claude will then find the CustomerInfo table and create a series of tests based on what makes sense in Collate, such as:

• Email format

• Phone number length

• Column count check

• Not null check for first and last name

• And more…

Those tests are then written back to Collate for that table.

[Video] How about Root Cause Analysis? This can be a challenging one to do manually. Consider the scenario where a test has failed; we need to find where it failed upstream and what impact it can have downstream. For this example, the column count test failed. Here is our Claude prompt:

“Can you do RCA for the table CustomerInfo”?

Here is what Claude came back with:

It found the problem and correctly diagnosed it. Identified the data flow pattern and also provided all the downstream entities that are impacted. Finally, it provided various recommendations. This is a simple example, but something simple can break a lot of things.

How can we make that RCA even more interesting? With a bit of out-of-the-box thinking, with this prompt:

“Create a dashboard view for me, summarizing the RCA, what might cause the failure and what we can do about it.”

This generates a convenient looking dashboard. Here is a snippet of it:

To wrap up, I’m going to give you some prompt ideas that you can adapt to your use cases:

1. [Video] Show me all datasets related to customer transactions from the last quarter

2. [Video] List all datasets tagged with ‘finance’ or ‘accounting’ glossary/tag

3. [Video] Generate a query to show monthly revenue trends for the past year

4. [Video] Create a SQL query to find our top 10 customers by purchase volume

5. [Video] Create a monthly customer acquisition and retention report

6. [Video] Build a customer retention dashboard based on the above (5) information. Use some sample data to build it for now; later, we can replace it with actual data from the above queries.

7. [Video] Can you help me find information about customers in my tables?

8. [Video] Put that information (7) in a modern user interface where each database has its own color and has a dynamically collapsible list of the data.

Summary

I hope this blog was able to get you up to speed on the incredible MCP additions in Collate v1.8 and spark your imagination on what might be possible. If you aren’t sure, just try it and see what you get.

Ready to get started? Sign up for Collate Free Tier of our managed OpenMetadata Service, or visit the Product Sandbox to try out Collate with demo data.