Introduction
Last week I published a video about building AI tools. Three minutes later, I had a perfectly formatted blog post on my website that I never actually wrote.
No, I didn't hire a virtual assistant or use some sketchy content mill. I built an automation that watches my YouTube channel, grabs the transcript from new videos, and uses Claude to turn them into blog posts that actually sound like me. Then it pushes everything directly to my website without me touching a single line of code.
The whole thing runs inside Claude Desktop using MCP servers, and honestly? What used to be this nagging task on my to-do list now just happens automatically. Plus, the blog posts are actually good, not some robotic AI nonsense (or maybe they are robotic nonsense? Claude wrote everything up to this point for this here very blog post, so I don't know, you be the judge).
Here's exactly how I built it, and why you might want to steal this idea for your own content workflow.
What This Automation Actually Does
Before we dive into the setup, let me paint you a picture of what this looks like in action:
- Automatic Video Detection: A Make.com scenario monitors your YouTube channel for new uploads
- Transcript Extraction: The system pulls the video transcript using Apify's YouTube scraper
- Smart Blog Creation: Claude analyzes the transcript and creates a structured blog post based on your writing style
- Direct Publishing: The formatted blog post gets pushed directly to your GitHub repository and goes live on your site
The whole process takes about 2-3 minutes from start to finish. And because it's all happening within Claude Desktop with your custom project knowledge, the output actually maintains your voice and style.
The Tech Stack You'll Need
This setup combines several tools that work together seamlessly:
- Make.com: Handles the YouTube monitoring and transcript extraction
- Apify: Scrapes YouTube for video transcripts (way more reliable than trying to build your own)
- Claude Desktop: Processes everything and handles the blog post creation
- GitHub MCP Server: Pushes the final blog post to your repository
- Google Drive: Stores the extracted transcripts for Claude to access
Don't worry if this sounds complex—I'll walk you through each piece step by step.
First things first: we need to set up the automation that monitors your YouTube channel and extracts transcripts. I originally tried building this in n8n, but their MCP server trigger is completely broken as of version 1.94. Make.com, on the other hand, works like a charm.
Getting Your Make.com MCP Token
Start by logging into Make.com and heading to your profile. You'll find a section called "API/MCP Access" where you can generate a new MCP token. This is what allows Claude to trigger your automations.
Connecting Make.com to Claude Desktop
Once you have your token, you'll need to add the Make.com MCP server to Claude Desktop:
- Go to Claude Desktop settings → Developer → Edit Config
- Add the Make.com MCP server configuration
- Insert your MCP token
- Restart Claude Desktop
The Make.com documentation makes this super straightforward—they provide the exact JSON configuration you need.
Building the YouTube Monitoring Scenario
Now for the fun part: creating the actual automation scenario. Here's how to set it up:
Step 1: YouTube Module
- Add a "Watch Videos in Channel" module
- Connect your YouTube account
- Set it to monitor your specific channel
- Limit to 1 video for testing

Step 2: Extract Video URL
- Add a "Set Variable" module
- Combine the base YouTube URL with the dynamic video ID
- This gives you the full YouTube URL for the next step

Step 3: Apify YouTube Scraper
- Connect to Apify (you'll need an account)
- Use the "YouTube Scraper from API Dojo" actor
- Set it to run synchronously (gives it time to process)
- Pass in your video URL from the previous step



This scraper handles the difficulty of extracting YouTube transcripts, which I found out the hard way was actually more difficult than I thought when I tried building my own transcript downloader. It costs about 50 cents per thousand videos scraped and saves you from all the headaches.

Processing and Cleaning the Transcript
The raw transcript data comes back in a pretty gnarly XML format with weird encoding and timestamps. We need to clean this up before sending it to Claude.
Step 4: HTTP Request
- Extract the actual transcript content from the Apify response
- This gets you the raw XML file with all the transcript text


Step 5-7: Text Processing
- Use Make.com's text parser modules to clean up the XML
- Remove the weird encoding (like ' for apostrophes)
- Aggregate everything into clean, readable text


I'll be honest—I had to ask ChatGPT for help with the regex patterns here, and I suggest you do the same. It's super easy. Just ask ChatGPT directly with something like: "help me clean up this XML. What do I tell make.com?" Text processing isn't my strong suit, but the important thing is that it works consistently.
Step 8: Thumbnail Extraction
- Grab the highest resolution thumbnail from the video
- This becomes the header image for your blog post

Step 9: Google Doc Creation
- Create a formatted Google Doc with the video title, thumbnail URL, and clean transcript
- This is what Claude will read to create the blog post

Training Claude on Your Writing Style
Here's where the magic happens. This isn't just about connecting tools—it's about creating a system that actually writes in your voice.
Setting Up a Claude Project
Create a dedicated Claude project and feed it everything you've ever written:
- Blog posts
- YouTube scripts
- Email newsletters
- Social media posts
- Any other writing samples
The more content you provide, the better Claude gets at mimicking your style. I've uploaded probably 50+ pieces of my writing, and the results are genuinely impressive.
Creating the Automation Prompt
Within your project, create a detailed prompt that outlines the entire workflow. Mine is honestly like 11 pages long, but it covers everything:
- How to access the Google Doc with the transcript
- What format to use for the blog post
- Specific guidelines about my writing style
- Internal linking strategies
- Markdown formatting requirements

The beauty of building this kind of automated workflow is that once you get it dialed in, it becomes incredibly consistent.
Setting Up GitHub Integration
The final piece is getting your blog post live on your website. If you're using a static site generator (like most modern sites), this means pushing to GitHub.
GitHub MCP Server Setup
Add the GitHub MCP server to your Claude Desktop configuration. If you need help with this setup, I've got a detailed tutorial on using Claude with MCP for publishing. Once it's connected, Claude can:
- Search your repositories
- Create new files
- Push changes
- Handle all the Git operations automatically
Blog Post Structure
Make sure your blog post template includes all the necessary frontmatter:
- SEO metadata
- Category and tags
- Publication date
- Featured image (your YouTube thumbnail)
- Author information
Claude handles all of this automatically based on the template in your project instructions.
The Human-in-the-Loop Advantage
What I love most about this system is that it's not completely automated. Claude presents you with the finished blog post before publishing, so you can:
- Review the content for accuracy
- Make style adjustments
- Add additional context or examples
- Ensure it matches your brand voice
This is where AI automation really shines—it handles the heavy lifting while keeping you in control of the final output.
Testing and Troubleshooting
Before you start relying on this system, run through it manually a few times:
- Test the Make.com scenario with an older video
- Verify the transcript extraction is working properly
- Check that Claude can access the Google Doc
- Make sure the GitHub integration pushes to the right location
- Confirm your website deploys the new blog post correctly
Most issues come from incorrect MCP server configurations or permission problems, so double-check those first.
Final Thoughts
What's so cool about MCP servers is how they let Claude talk to basically anything. I especially love a workflow like this because my lizard brain cannot physically withstand multiple applications running at the same time. I need one interface, or as few interfaces as possible, and Claude Desktop with MCP servers makes this possible.
It's exciting to see where this technology can go and how we can start creating things with natural language. It helps bring down the friction of producing content and just lets you get your ideas out there quicker. That's what excites me about this automation and MCP servers in general.