No More Writing SQL for Quick Analysis

0:00Hello data nerd. In this video, you'll learn how to quickly get insights on any data using motherduct. The only thing you need to know is SQL. I'm just kidding. AI will actually do that for you. So, you just need to speak English or if you're more comfortable.

0:27So, we'll be able to ask natural language questions in our AI tool and get back results thanks to the Motoc MCP server. MCP in a nutshell is a standard for LMS to interact with external tools.

0:40Thinks of the MCP protocol like an USB interface. You can connect to any MCP server to your LM which acts as an MCP client and can perform action defined in that MCP. For us, it is query a database. We'll do a deep dive on MCP on this channel, but that's for another video. So yeah, you want to subscribe to

1:00this channel to not miss it up. So to set up your MCP client, you usually have two options in your LLM tool, whether it's your chat GPT cloud cursor or others. The option one is an improved remote MCP server and the setup is just a few clicks. The option two, you run the MCP server locally. In this video,

1:19we'll go with option one since it's the easiest route is true code where the model NCP server is already approved.

1:27And at the time of recording this video, we are in the process of getting verified on other popular platform as well. So check the docs in the description if you want to know the latest support on that topic. So on cloud desktop you can add the MCP server to settings to connectors browse connector and then you can search for

1:46the modak uh MCP connector. You're going to be prompt to go to a web authentification. Uh if you don't have an account yet on modak you can create one for free and then you're going to be prompt to open the cloud app again and this configuration will be done. And as you can see if we go back to the

2:08settings now I can see all the reaction that my MCP server can do. I can ask actually doc question this column list database and the most important one uh query uh of course uh the data on the

2:25right side is that I can change u how the LLM will ask for my permission or not before running an action. If you allow everything by default, the loop is faster because if a lm does an error, it's not going to prompt again for confirming a specific action. It'll just do it directly. But it also has more

2:47risk. But I'm going to activate the yolo mode here because anyway, the MCP server can only do read only SQL query. So there's no real danger. All right. So now let's confirm that uh the uh MCP is actually working and we're going to ask can you list my modd DBs really simple action. So you see it's done uh an

3:08action on a list database and he can list all the database I have would it be a mo DB share or a dlick uh database. So

3:18now we are all right to continue and start asking question in plain English. So first give me some analytics on the I

3:28can use database. So I on purpose give it really a broad question. Of course the more accurate and narrower your prompt is the better the results. But it's just to show you that for simple exploratory data analysis this is really

3:46really good. you see is getting the columns uh getting the schema and then performing relevant queries for analytics. So now I have more time I guess to talk with you our review. All right. So now it's getting me some information. You see it's pretty quite already a large table.

4:05We have more than 40 millions rows uh the date range. So he's done some minmax around just to get you know a vision of uh just to get an idea of the shape of the data and uh basically the number of comments and of course some analytics on the top most uh popular uh stories and you see you have the relevant years and

4:27then other analytics like the most active users and so forth. That's for the first use case. For a second use case we're going to query a file on a object storage. So not stored on motherdox typically on AWS S3 a park file and for that it's a private bucket.

4:45So the thing first that I'm going to do is log to using the AWS uh CLI to my SSO

4:53credential. And when it's done I can launch any local ductb client but the easiest one is a dagb uh cli client. I'm

5:03going to connect to motoduck using the attach nd command. So now I'm authenticate to motodak to the dagbli and then create uh a secrets in motoc.

5:14I'm refreshing my AWS credential in modc. You can also go to the web UI and uh head over the secrets tab and then uh

5:25add manually uh this one. So now I can ask you give me some analytics on that

5:32file and I'm going to copy paste my private uh file in a private history bucket and because motoc has access to

5:42uh this bucket I'm leveraging the cloud network so there is nothing transiting actually to my laptop and of course the speeds of the bandwidth to query would it be a large file or not is faster and As you can see there is sometimes some error but just the LMS is iterating and

6:03fixing based on the error it's getting. So there is really no copy paste for me to just say hey this query fail uh try something else. It's doing that automatically for me. So now I can see that the movie park file is actually a data set which is a vector ready data set good for semantic search or

6:25recommendation system. And the LM is asking me actually do you want to u do you want to uh run some sample vector similarity search.

6:36So let's let's actually do that. Now it's going to query using some specific function of Doug DB to do uh a sim vector similarity search. And again as a result I have a couple of information about similar movie to Toy Story or uh Batman as you can see with the similarity score. All right. For a third example, we're going to push the more

6:58forward and try to query actually a public API with again just mo duct DB

7:05behind the scene because it can query an HTTP endpoint and it can parse a response from an API, a JSON response pretty really well. So I'm going to ask use modak mimcp to query some common

7:21data repos to see what's the top language programming language used for instance uh I'm giving an example of the github

7:33endpoints for the ddb repo so this is returning this endpoint the percentage in bytes or what's the the amounts of a typical programming language of a given repo and so you see it's going to go through multiple repository I haven't asked actually any which repository it should go just again a wide broad question data repository and now it's start to give me a nice uh

8:01top language per repository and you see we have trino click spark and see what's the top language used in each of those and we've uh given actually a specific conclusion ion and key insights which is C++ dominates the data infrastructure worldb click house and arrow are all based on C++ R is rising Python is everywhere but but but mostly as binding

8:28test not for core engine and the GVM still matters um I have to disagree with that but that's for another video that's it for this video we saw how easy it is to set up the MCP motoduck uh server using code and we didn't do any SQL the

8:47LM just interacts with himself loops if there is error or suggest things and we can get some first insight really quickly with really broad question I didn't push the prompts at all if you want to see other AI workflow like this let me know in the comment what you would like to see and in [music] the meantime take care of your ducks and

9:08I'll see you in the next one cheers