# Build a MCP Server of Your Favorite YouTuber with Claude + Cursor ## Metadata - **Published:** 6/19/2025 - **Duration:** 9 minutes - **YouTube URL:** https://youtube.com/watch?v=PG5RSX3jXP0 - **Channel:** nerding.io ## Description Join the Community: https://nas.io/vibe-coding-retreat In this video, we show how to create your own MCP server that mimics your favorite YouTuber—powered by Claude and developed in Cursor using the Model Context Protocol (MCP). This setup lets you spin up an AI persona that: 🎙️ Talks like your favorite creator 📺 Responds with insights based on past content 🧠 Leverages Claude’s reasoning to act like them ⚙️ Is fully accessible by any MCP-compatible client 🛠️ What You’ll Learn: How to model an AI agent after a specific YouTuber’s tone, style, and knowledge - Using Claude to power agent responses - Setting up a local MCP server with custom tools - Building in Cursor IDE for faster dev and testing - Creating prompt templates and memory using transcripts or video metadata 💡 Use Cases: - AI assistants that mirror thought leaders - Fan-powered Q&A bots - Training agents on niche content for edutainment or support - Prototype AI-driven creator collabs 🔗 Resources & Links: 🧠 Cursor IDE: https://cursor.so 🤖 Claude: https://claude.ai 📩 Newsletter: https://sendfox.com/nerdingio 📞 Book a Call: https://calendar.app.google/M1iU6X2x18metzDeA 💬 Which creator would you turn into an AI agent? 👍 Like & Subscribe to see more weird + powerful AI experiments! ## Key Highlights ### 1. Automated YouTube Vector Store Creation The video demonstrates building an automated system that pulls transcripts from a YouTuber's videos, creates vector embeddings, and stores them for conversational interaction. ### 2. MCP Server for Personalized Knowledge Access The presenter builds a MCP server to access information from favorite YouTubers, allowing querying and summarization of video content for practical application. ### 3. Workflow Generation from YouTube Content The system translates insights from the YouTube video transcripts into executable N8N workflows, automating complex tasks described in the content. ### 4. Integrating with Claude and Cursor The video showcases how the created MCP server and vector store can be accessed and utilized within coding environments like Cursor and Claude, enabling AI-assisted workflows. ### 5. Cron Job for Daily Content Embedding A cron job is used to automate the process of checking for new videos daily, extracting transcripts, and updating the vector store for continuous learning. ## Summary ## Video Summary: Build a MCP Server of Your Favorite YouTuber with Claude + Cursor **1. Executive Summary:** This video demonstrates how to build a personalized AI assistant that mimics a specific YouTuber by creating a vector store from their video transcripts and leveraging Claude's AI capabilities via an MCP server. The system allows users to interact with the AI persona in coding environments like Cursor IDE to gain insights, generate workflows, and automate tasks based on the YouTuber's content. **2. Main Topics Covered:** * **Automated YouTube Vector Store Creation:** Process for extracting transcripts from YouTube videos and creating vector embeddings for conversational AI. * **MCP Server Setup:** Building and configuring an MCP server to access and query the created vector store, enabling personalized knowledge access from your favorite YouTubers. * **Workflow Generation:** Translating insights from YouTube video transcripts into executable workflows using tools like N8N. * **Integration with Claude and Cursor IDE:** Utilizing the MCP server and vector store within coding environments like Claude and Cursor for AI-assisted workflows. * **Automation with Cron Jobs:** Implementing cron jobs to automate the daily process of checking for new videos, extracting transcripts, and updating the vector store. * **Custom Tool Building:** Using a custom tool to build an ai agent based on a content creator's tone, style, and knowledge. **3. Key Takeaways:** * You can create a personalized AI assistant by leveraging a YouTuber's content using vector embeddings and Claude AI. * An MCP server allows access to the vector store, enabling querying and summarization of video content. * The system can generate workflows from video content and integrate with coding environments like Cursor and Claude. * Automated cron jobs ensure the vector store is continuously updated with new content. * The process can be replicated for other YouTubers or content creators. **4. Notable Quotes or Examples:** * "I'm going to use a MCP server that I made, which will actually leverage uh every video that comes out from him, turn it into a vector store, and then allow me to chat with it." - JD * The video demonstrates an example where the AI agent is asked to use the "Boring Marketer" MCP server to break down video workflows and then create them in N8N. * The example showcased generating N8N workflows for "ad prediction," demonstrating the practical application of the generated knowledge. **5. Target Audience:** * Software developers * AI enthusiasts * Individuals interested in automating workflows * Content creators seeking to build AI assistants * Fans of specific YouTubers or thought leaders * People looking to integrate AI with their coding environment ## Full Transcript Hey everyone, welcome to Nerding.io. I'm JD and today what we're going to go through is how you can actually build your own YouTube vector store, which basically means every new video of your favorite YouTuber you can actually store and have a conversation with it. We're actually going to do this through NAN and MCP. So with that, let's go ahead and get started. All right, so we're going to go through this together. So, I really liked the boring marketer. Just came out with a video specifically talking about how to do uh award-winning ads. And I am terrible at ads and terrible at marketing in general. I'm trying to figure out exactly how to to do this. So, I'm going to use a MCP server that I made, which will actually leverage uh every video that comes out from him, turn it into a vector store, and then allow me to chat with it. So, we're going to go ahead and give this a try and I'll just kind of show you what I'm talking about. So basically I've come in here I have a uh MCP and now I am going to say uh use the B boring marketer MCP uh you know the title of the video break down the workflows but then create them for N and we'll just see what happens. All right. So, it's actually going out and searching. We're just going to go ahead and say always. Uh, and then it'll just keep searching. So, what it's doing is it's trying to go out and find information from the database of the transcript from the YouTube video. And it's collecting all this information. It's figuring out uh different information from from that transcript. Uh, And it looks like it's grabbing it from Justin's ad prediction workflow. Cool. That's exactly what I wanted to do. And it's pulling out the workflow breakdown. And then it's actually going out and building me the N8 workflow. Uh so this is really awesome. Don't know if it's totally going to work, but it's still cool that it's taking the context from each one of these and then building out uh multiple different NAN workflows. Should be noted that like in that video they they don't use NAD. Uh they use like a Viz Studio or Viz Studio or something. So um pretty cool that this is generating on the fly and that I can just like ask it questions. So, I'm gonna let this run and we'll see what we come back with and then you can ask some more questions. All right. So, we let this run and uh it had to hit a pause. So, go ahead and continue. But we can see that we actually have three different Oh, it's redoing the adding to this one. All right. So this finally finished uh literally from the ad prediction workflow breakdown into four components. It created the the prediction workflow in and it created a ad prediction batch workflow and then a prediction dashboard API uh which is super interesting. even gave us the instructions of how we can actually go build this database and put it together. Again, I'm doing this on the fly, so uh wasn't entirely sure what it was going to output. This is actually pretty sweet. So, if I want to take a look at the uh the N8N, all I need to do is is copy this. Um, and we'll pull this into to N8N and take a look. I do want to kind of show like this is how I'm actually serving the MCP servers. So I went through and I actually made a daily uh embedding of every video that comes out for boring marketer as well as others that I follow and a unified MCP server. But this is how you can actually serve these embeddings uh up. Real quick everyone, if you haven't already, please remember to like and subscribe. It helps more than you know. Also, if you're looking for hands-on training, there's a link down below for virtual and in-person retreats. With that, let's get back to it. So, yeah, let's uh let's create a new workflow and see what this what this is doing. So, we can just paste that in. So, this is pretty wild. This is uh looks pretty close. It is taking the web hook, setting the variables. Let's see if we got a prompt here. That would be kind of interesting. No prompt. So, we'd have to go back through and uh figure out the the prompt or like pick a different uh operation. Again, we have like our database connections. So, it still gave us like the the baseline for for what we would need to go build. I would probably clean this up a bit. Uh, and let's see what we've got with the batch. Same thing. We can just copy and paste. Oh, property. All right. So, looks like there's an issue in the ad. So, we'll skip add to this one and maybe tell it. Oh, looks like errors in both the other two. So, probably need to go back and refine those. But, uh, just to start still again, we've got this, uh, this workflow that came directly out of the the embedding from the YouTube uh, post. So, now I'm going to kind of go through how this has actually built. In order to do this, what I'm doing is I am taking a crown job and I'm saying I want this to check every day. I want to pick the YouTubers that I'm I'm connecting to on a regular basis. I am going to set that information. This is the big one. I'm actually going to write the string for how to what what videos to go back and fetch, right? So, I can't just say, "Oh, get the latest new video." You have to look back and see what's happened in the past day. And that's your check. Then, you're going to get the video ID. You're going to do an if statement. This if statement is basically to look and see if the uh if this is a short or not. Then, we're going to go ahead and process every single one of those. That if it is a long video, then go ahead and get the transcript. Uh, I had to do a rate limiting delay. Combine everything because this gives it to me in chunks. Combine everything together. Then go ahead and chunk that data back out into a vector store. Save all of that information. Uh, and this chat is really just for testing. So, you can actually open and uh do like a a test here. It's going to connect to OpenAI. So now you have the ability to do the exact same thing uh in OpenAI and ask it questions. So we could go back and ask the same thing. Probably probably not ask it to do everything in a n, but it'll still go out and pull back from the embedding. So right there we're actually pulling back information. So now you have the ability to uh it's pulling multiple times too. So you have the ability to to actually do this in N8. You have the ability to do it in quad actually have the ability to do this in cursor. So if we take that same URL uh for MCP and we go over to our MCP.json and we just say the exact same thing. In theory, this should work and we should be able to pull this information directly into our uh cursor again. Cool. Yeah, we're doing the search and now we're actually able to code as well as uh get information from the boring marketer uh as we're building. All right, everyone. That's it for us today. What we went through is how you can actually automate your own YouTube vector store as well as create an MCP chat in both cursor and claude as well as in any end. If you're interested in the workflows, please remember to sign up for the newsletter. And with that, happy nerding. --- *Generated for LLM consumption from nerding.io video library*