The MCP Series
In our previous posts we looked at what exactly MCP is and how it works behind the scene. Now it’s time to tie it all together and build some useful stuff with MCP.
Before we begin, I’m going to let you in on a little secret. I used Claude with MCP to help me write this series. Shocker!
Claude has always been my favourite for creating content. You can read more about how all the different AI chatbots and tools fit in my stack. Before MCP, I’d have to do a bunch of research on my own, then feed Claude all that context before working with it to craft a post.
That’s a lot of work so I decided to build my own AI agent to automate the research work too. I used Exa to find resources to include and sent that to Claude via the API. But the agent was a bit limited too, and I missed the chat experience in the Claude app, man and machine working together in harmony to write a blog post.
Now, with MCP, I can do exactly that. This is the first of many examples of building AI workflows with MCP I’m going to be sharing with you today. Let’s dive in.
Get more deep dives on AI
Like this post? Sign up for my newsletter and get notified every time I do a deep dive like this one.
1. From Chat to Blog Post
I’m not a fan of telling the AI to write a blog post and then directly publishing it without any human oversight or input. That’s just AI slop.
Plus, I write because I want my thoughts and opinions to be read and critiqued by others. I get no satisfaction and will not learn if an AI does everything for me.
So I use AI (Claude specifically) as my writing partner. It’s not a linear process. I start with some thesis or topic I want to explore (like this MCP series) and use Claude to help me create an outline.
Once I’m satisfied with the outline, I do some research to back up what I want to say. Exa has an official MCP server that I’ve installed, allowing Claude to search the web and pull interesting content.
I use Exa for a number of reasons that I’ve listed in my Guide to Exa, the primary one being it tends to pull in much deeper and richer content, including research papers, which is great for writing technical content.
I also use the Brave MCP and a custom Twitter MCP I built for myself. Brave gets me more recent news type results and Twitter pulls in trending conversations about said topic.
After Claude has pulled these results, I read through them and may tweak the outline again based on the new information.
At this point, I’m ready to write the actual article. I have a thesis or topic, a structure to present said topic, and real data or resources to support me.
The process of writing is an even longer back and forth. I ask Claude to flesh out the first chapter, then I make edits, and we go back and forth until I’m satisfied.
I’ll often ask Claude to poke holes in it or provide counter-arguments. I’ll also try to include the research or even ask Claude to pull in more specific research for this chapter.
I repeat this for every chapter until the whole article is written. At this point, we have a fully written out articles, with a clean structure and headers, and links to supporting data or resources.
Finally, with the WP MCP server (I built one for myself), I ask Claude to post it. It first asks for my permission, and then in pushes the content to my WordPress site (this blog you’re reading) in draft status. I then log in to WP, make any final tweaks and edits, and hit publish.
And that, dear reader, is how I wrote these very words you’re reading right now!
For a full video overview, watch this:
2. From Chat to GitHub Repo
Quite similar to the way you can chat with Claude to create WordPress blog posts, you can also create GitHub repositories using the GitHub MCP.
Me: What changed in our authentication system between the last release and now?
Claude: [Using Git MCP] I see 7 commits affecting authentication since v2.1.0. The most significant changes include:
1. Migration from JWT to OAuth2 for API authentication
2. Implementation of refresh token rotation
3. Addition of 2FA for admin accounts
The primary developer was @jsmith, and all changes passed CI. Would you like me to explain any specific change in detail?This isn’t just a demo—it’s how developers are working right now. As shown in “This Claude MCP GitHub AI AGENT Changes EVERYTHING!“, MCP is revolutionizing how developers interact with their repositories.
3. End to End Sales Agent
This is a pretty cool workflow built by Robert Lukoszko where he has an MCP server that connects to his LinkedIn. For anyone who comments under a post, it’s able to extract the profile of that person and go and do some research on that person to find out more about them as well as their email. He also has an MCP server connected to his email so he can then have an email sent to the people who commented on that post.
Essentially a very powerful sales agent that automatically goes through the comments on any LinkedIn post and then reaches out to them via email.
4. Customer Support Automation
Karan Vaidya, from Composio, shared a couple of really cool customer support automations that he built out with Cursor and MCPs.
The first one fetches customer support emails, analyzes the issue, solves the issue, and then responds to the email with the correct solution.
If it can’t correct the solution, it uses Git Blame to find the responsible dev, creates a ticket on Linear, and then notifies the developer on Slack about the issue.
5. Other Creative Applications
Beyond the expected use cases, I’ve been amazed by the creative ways people are applying MCP technology.
Personal Knowledge Management
David Teren’s Claude Server MCP transforms knowledge management with persistent context across conversation sessions:
Me: What connections exist between my project notes on renewable energy and my reading notes on urban planning?
Claude: [Using Claude Server MCP] I found 7 conceptual links between your renewable energy and urban planning notes. The strongest connections involve microgrids, community solar projects, and transit-oriented development. You explored similar funding mechanisms in both areas but haven't explicitly connected them in your notes. Would you like me to create a new note mapping these relationships?Advanced Document Research
Alexander Komyagin built an Agentic RAG solution using MCP to fundamentally change how he interacts with documents (read his article here):
Me: What do our financial projections say about Q3 revenue growth compared to analyst expectations?
Claude: [Using Agentic RAG MCP] Let me examine your documents...
After analyzing your Q3 projections and comparing with analyst reports from Bloomberg and Reuters, I've found a 5.7% discrepancy. Your internal projections show 12.3% growth while consensus estimates are 18%. The primary factors driving this difference appear to be more conservative supply chain recovery assumptions in your models.Instead of the traditional static retrieval systems, his MCP implementation lets Claude autonomously decide its search strategy and combine information from multiple document sources—just like a skilled researcher would.
Automated 3D Designs
Siddharth Ahuja connected the Blender MCP server to Claude to create 3D designs and scenes with just a few prompts.
Learning Helper
Giacomo connected Claude to Rember, a tool to create flashcards, to help him automatically create flashcards to remember important concepts while chatting with Claude.
What Will You Build?
What makes MCP revolutionary isn’t just the technology itself, it’s the connections it enables. Right now, only Claude, Cursor and a handful of other AI companies have MCP clients that enable you to extend their functionality by adding MCP servers.
But imagine what happens as more companies build out MCP clients. The ability to seamlessly integrate AI into your existing workflows, tools, and data sources creates a multiplicative effect that transforms productivity.
As you explore MCP for yourself, start with a simple question: “What tasks take up most of my time but don’t require my unique human judgment?” Those are perfect candidates for automation.
The real power comes when you combine multiple MCP servers. Need to analyze data, create visualizations, and publish findings to your blog? Chain together database, visualization, and WordPress MCPs for a workflow that would have been science fiction just months ago.
And if you need help building these out, drop your email below and I’ll reach out.
Get more deep dives on AI
Like this post? Sign up for my newsletter and get notified every time I do a deep dive like this one.