Build an AI-Powered Job Search Agent | Bright Data MCP + Nebius AI Studio

0:00 Hey there everyone, I am Arindam and today I'm going to show you how you can build an job searching agent using bright data MCP server. So we will create an streamllet app where you can directly interact with this agents and I will also show you how you can run this locally and how to implement this. So first let's see this in action. So we will paste our LinkedIn profile URL and the Navius API keys. You can get it from studio.navs.com. You just have to click here get API keys and you will create an So we can just create a demo and create this and we'll copy this and paste it here. Awesome. Next we need to paste our LinkedIn profile URL. Now we will analyze the profile and it will start running. Awesome. So as you can see here it has generated the response. So here is a small profile summary of mine and some top skills and it has suggested some roles content creators and now it has given me the job results and you can see it has given me a lot of details like this full-time location. It has also given me the apply link. So you can just click here and you can get to the live link and you can just apply here.

1:16 So now let's see how this works. So as we have seen before first we need to give the LinkedIn profile of the user and our LinkedIn profile analyzer agent it will use the bride data MCP server and get the data of that LinkedIn profile and based on the analyze profile information it passes that to the next agent which is our job suggestions agent and this agent takes the information and decides which domain it is suitable for and based on that domain we generate a URL for the job board and we pass that to our next agent which is our job search agent. So our job search agent then again uses the bright data MCP server to get the details of the job and it gets the data from that MCP tool and it passes everything to our final agent which is the summary agent and the summary agent then takes all the information and creates a detailed report summarizing all the information jobs and suggestions. For the text stack, we have used the bride data MCP server and the open agents SDK for multi- aent orchestration and navi studio for LLMs. So you can check out the bride data MCP server. With this you can easily give web access to your LLMs.

2:31 And if you're building agents or using any MCP clients that say cursor plot, it is super useful to get realtime information as you can also see in this diagram. And this is really easy to use. You can check the quick start and they have a lot of available tools. You can see they have almost 30 plus tools that you can use for your agents. You can use them based on your use case. And for the LMS part, we have used NBS AI studio. NBS AI studio provides state ofart opensource inference models. So you can get textto text model, embedding models, text to image model, vision models. So you can use them accordingly.

3:07 Now we will see how we can build your job searching git. So for that you need to create an account on bright data. So you can go to bright data.com and get started for free. I have already created an account. So I will just continue with Google and you will go to a dashboard like this. So and now you go to proxy and scraping. So here you'll also see that I have already created web unlocker API and browser API. So you might not get this thing as I have already used this. So have to do is you have to go to create a browser API. So just go to here create a zone. You can add a zone name here and you can add some advanced settings as well. But for this example you will not need those. So you can just add this and you will get one browser API added. So I will skip this.

4:04 Similarly for the web and locker API you can go to here and create a zone and similarly you just add a zone name and add here and you will get everything done. Awesome. So as you can see here when you click here you'll get the direct API access and the API key for our project we'll be needing this API key. So we need to copy this and we are ready to start working on our project. We have created a bright data account. We have our LLM's API keys ready. So let's go to our ID. So this is our entry point of our file. So we have the app.py file, the job agents file and the MCP server file. So let's understand one by one. So let's go to our job agents file. So here we are creating all the agents that I have already shown you in the diagram. So first we are importing the modules and then we are getting the API keys, the base URL and defining the client. If you have seen my previous video using Openia agent SDK, you know how this works and in case you haven't checked that you can check that out. I will add the links and first we are creating the LinkedIn agent. LinkedIn agent will use the bright data MCP server and get the details of the LinkedIn profile. You can see the instruction here. It analyzes the profile and gets these details. And here we are using that metal lama model.

5:25 So next up it passes the details to the job suggestion agent. So here you can see that it is not using any MCP server because we don't need to access any external data. So we are just getting the data from the previous agent that is our LinkedIn agent. And here we are based on that agent we are just defining which domain the user is perfect for and passing it to the next agent which is an URL generator agent. So the URL data agent basically is getting the job board URL. We are again not using any MCP server just because we don't need to access any external data here. We are just passing the URL to the next agent which is our job search agent. So here you can see we are using the MCP server because we will be using to get the details of the job boards and the job applications. So you can see here we have mentioned all the steps to take the URL from the previous agent and then provide the URL to the job listings and get these details that you have seen in the document and get the following details the company name, job title, job type, location and apply URL and also we have told it to format it in such way.

6:39 Finally I have added an extra agent because when you scape the data you don't get the exact job URL. to get the job URL I have added this thing where it creates a base URL and based on the job id from that link it just creates a simple URL that doesn't needs any sign in to view the URL so to simplify the job link process I have created this URL parser agent so you can skip this if you want cool finally we are passing every details to our summary agent summary agent is taking all the details and creating a summary of the profile summary your top skills and some suggested roles based on the information that it has got job links and everything that we have shared. So now we are creating the workflow here. So as you can see here we are using the starting agent as the LinkedIn agent and passing the input as the input query and we are getting the result as the LinkedIn result. so here we're getting the profile analysis that you have seen in the diagram. Next up we'll be using the job suggestions agent. So one thing to keep in mind here we are passing the output of the previous agent as a input.

7:49 So as you can see you are passing the LinkedIn result that we have got in the previous section and passing it as input here and then we are getting the result of this. Similarly, we are also passing this result as the input of the next agent which is our URL generator agent and we are getting that job link and we are passing the job link here as the input for the job search agent and we are getting the job search result and similarly you know we are doing this next step as well and finally we are passing the details to our summary agent and this is the input I have created for this and we are passing that's input here and the starting agent as summary agent and we are getting the final result which is the summary result and we are just returning that awesome now let's go to our MCP server so in this file we are initializing our MCP server so first we are just importing the modules and just creating a global variable here and then we are defining the server with the params that we have gotten from the GitHub repo and also passing the APIs that you have already created. You just need to go to your env file and you can just paste your information. The bright API key, the browser o, the navs API key and we're good to go. And for the browser o go to this browser API that we have created and here you will find the API key. So you just copy this to the clipboard and paste it. That's it. Awesome. Now let's go to our MCB file. So you can keep the web unlocker zone as MCP unlocker. So this is something they do by default. In case you want to change this, let's say you want to put it as MCP unlocker one that we have created, you can just mention that. Otherwise, you can skip this as well. This is completely optional. And finally, we are creating the MCP server and returning this using that get MCP server function. Now let's go to our app.py.

9:52 Similarly, we are just importing all those modules and the functions that we have created in the other files. The run analysis function, the MCB server function. And here we are just creating the UI. So if you have ever used frame rate, you know how this works. So here we created the analyze profile function. So first it waits for the MCP server to initialize and then it calls the function with the MCP server and the LinkedIn URL. Next we go to our main function where we are just importing the UI elements and creating this sidebar. So we are just taking the LinkedIn URL and the NBS API key here and we are just calling this function and that's it. Sounds easy, right? So now let's just run this. So to run this I have already created a virtual environment. You can also create yours and to run this just type streamllet run.

10:46 and it will create an UI like this and you can start using this. You can also deploy this as well. Awesome. So we saw how to create an job search agent using bright data MCP server and I have already made the code public. You can go to the GitHub URL in the description and you will find all the instruction to run this locally and also I have embedded a GIF for your better understanding. You can see the project structure installation. You just need to clone this, create a virtual environment and install the requirements and run this as you have already seen here. That's it for today.

11:21 I hope you like this. In case you have any doubts, feel free to write them in the comments. And thanks a lot for watching this video till the end. I will see you in the next video. Bite.