The Ultimate Guide to Model Context Protocol, Part 1: What is MCP

Well hello there! I take it you’ve been hearing about this MCP business online and have meandered over to my humble website looking for answers. This post shall shed some light on the entire affair.

Today’s AI assistants like Claude, Grok, and ChatGPT, are a clever lot, brimming with facts and ready to toss out answers faster than you can say “what ho!” to any query you lob their way. But when it comes to rolling up the sleeves and actually doing something for you, well, they fall short.

It’s like having a butler who’s all ears and sage nods, but when you cry, “Sort out my blasted emails!” or “Fish up those receipts before the taxman comes calling!” he merely blinks and offers a sympathetic, “Quite so, sir,” without lifting a finger.

You’d get rid of him really fast.

That’s where the Model Context Protocol comes in. Developed and open-sourced by Anthropic in November, 2024, MCP is a new standard for your AI butler to connect to your data, or any other siloed data source, and take actions, in a secure manner.

In this post, the first of a series on MCP, we’ll cover what it is, why it’s different from an API call or integration, and how you can get started with using it in just a few minutes.

Sign up if you want to know when I release the next post in the series.

What is MCP?

The Model Context Protocol (MCP) is like a universal translator between AI models and your digital world. Just as USB-C provides a standardized way to connect your devices to various accessories, MCP provides a standardized way for AI to securely access and work with your files, apps, and online services.

Don’t worry about the technical jargon—here’s what you need to know:

MCP Host: The application where you interact with AI (like Claude Desktop). Think of this as the “home base” where you chat with your AI assistant.

MCP Server: A special program that gives AI access to specific resources (like your files or Slack). Each server is like a specialized tour guide that knows one area extremely well.

MCP Client: The behind-the-scenes connector that lets the host talk to servers. You don’t need to worry about this part—it works automatically.

How Is It Different from an API or Integration?

Ok so essentially MCP is a way for Claude to talk to your data or some external service. Isn’t that literally what an API or an integration does? Why are we complicating this?

Well, first of all, MCP sounds cooler than API.

But yes, you could do this with an API call, except it’s complicated. For starters, you’d need to know how to code and make an API call. Then you’d need to configure Claude or another AI assistant to actually make that API call. And then you’d need to repeat that for everything you want it to access – your files, your email, your Slack, and so on. Exhausting, what?

Why doesn’t Anthropic just build integrations to all these apps instead? Well, again, that’s a lot of work. So they’ve basically just pawned off all that work to the developer community to build MCP servers.

It’s a bit of a middle ground, but still very simple for the end user. You find an MCP server by a third party that does the thing you want it to do, you tell Claude to use that MCP server, and Bob’s your uncle, job done.

Can’t find an MCP server? Make your own (we’ll get to how in a later post in this series).

In fact, some MCP servers are actually just wrappers over an API! But there are additional benefits:

1. Standardized Security and Control
   • MCP servers enforce strict access rules, requiring user approval for actions (e.g., a tool like write-file needs explicit consent). APIs, by contrast, rely on developers to implement security, which can vary widely.
   • Example: An MCP server accessing your Slack channels ensures the AI only reads what you allow, unlike an API token that might grant full access if not scoped properly.
2. Two-Way Communication
   • MCP supports bidirectional data flow, enabling AI models to not just fetch data but also act on it. For instance, an MCP File System server can let an AI read a document, summarize it, and save the summary back—all within one protocol.
   • APIs typically require separate calls for each step, increasing complexity.
3. AI-Specific Optimization
   • MCP provides “tools” (callable functions) and “prompts” (pre-written templates) that align with how AI models process information. For example, a weather MCP server might offer a get-forecast tool that returns data in a format an AI can easily digest, reducing preprocessing.
   • APIs deliver raw data (e.g., JSON), leaving it to developers to adapt it for AI use.
4. Local and Remote Flexibility
   • MCP servers can connect to local resources, like your computer’s file system, or remote like a Chrome browser without needing a web-based API.
   • Example: The Puppeteer MCP server controls a browser locally, while a Google Maps MCP server hits a remote API, blending both worlds.
5. Simplified Integration
   • MCP standardizes how AI models interact with external systems, reducing the need for custom code per API. A developer can use one MCP client to connect to multiple servers (e.g., Slack, GitHub), whereas APIs require unique integrations for each.

Practical Scenarios: API vs. MCP

Do You Need MCP?

• If you’re just fetching data: Stick with APIs—they’re simpler for basic tasks like grabbing stock prices.
• If you’re powering an AI: MCP shines when you need your AI to interact with the world, locally or remotely, in a secure, controlled way. For example, integrating Claude with your file system via MCP is safer and easier than building an API for it.

What MCP Can Do For You: Real-World Examples

Ok, hopefully I’ve convinced you that an MCP is actually useful and not just Silicon Valley reinventing something that already exists.

Now, let’s look at some real world examples:

Personal Productivity

• File Organization: “Claude, can you organize my downloaded files into folders by type and date?” With MCP, Claude can actually do this for you, not just tell you how, while you polish off your second donut of the morning.
• Email Management: “Summarize all my unread emails from the bigwigs,” you plead. With MCP, Claude dives into the inbox, sifts through the missives, and delivers a pithy précis, perhaps even firing it off via Slack or a text.
• Note Analysis: “Claude, cast an eye over my meeting notes from the past month and whip up an action plan, there’s a good chap.” With MCP, Claude rummages through your scribblings, plucks out the juicy bits, and adds todos to your task management app faster than you can blink.

Information Access

• Document Search: “Find me the skinny on our budget projections in my documents folder,” you command. With MCP, Claude ferrets through your private stash without so much as a whisper to the internet, emerging triumphant with the goods, like a bloodhound on the scent.
• PDF Q&A: “What were the key recommendations in that report I nabbed yesterday?” you muse. Claude, armed with MCP, tracks down the PDF, pores over it like a don at his books, and serves up the answers with the precision of a well-aimed dart.

Communication

• Message Drafting: “Draft a Slack message to the troops summing up the quarterly report on my desktop,” you say. With MCP, Claude saunters over to your files, has a butcher’s at the report, and taps out a message with the finesse of a seasoned clubman penning a note to the committee.
• Conversation Summaries: “What were the main thrusts of yesterday’s team chitchat?” you ask. Claude, with MCP as trusty steed, gallops through the chat logs and returns with a tidy summary, sparing you the bother of wading through the blather yourself.

Web Search

• Browse the internet: “Dig up the latest gossip on AI and give me the lowdown,” you request. With the Brave or Exa MCP servers, Claude scours the web and delivers a crisp rundown like a newsboy hawking the evening edition.
• Find restaurants: “What are the top Thai eateries near my digs?” you wonder aloud. With the Google Maps MCP server in play, Claude not only unearths the finest curry houses but pops the addresses your way, like a cabbie with a knack for spice.

By the way, I’m using Claude as an example, but any company can become an MCP Host and create their own client. This instantly opens up a world of possibilities for their users.

Cursor, for example, also built an MCP client. So, just like with Claude, you can install a web scraper MCP and have Cursor scrape the most up to date documentation for a Python package to use in the code it generates.

Top MCP Servers and What They Do

MCP servers are the building blocks that give AI access to specific parts of your digital world. Here are the most popular ones:

1. File System MCP Server: Lets AI safely work with files and folders on your computer
2. Slack MCP Server: Enables Claude to post messages, reply to threads, add reactions, and more in Slack.
3. GitHub MCP Server: Helps manage code repositories and issues.
4. Google Maps MCP Server: Enables location-based assistance
5. Brave MCP Server: An MCP server implementation that integrates the Brave Search API, providing both web and local search capabilities.

You can find a list of servers on the MCP site. Each server handles one specific type of connection, and you can install exactly the ones you need.

Getting Started in 10 Minutes

Ready to try MCP yourself? Here’s how to get started:

1. Download Claude for Desktop

Right now, MCP servers are hosted locally (on your computer), so we need a local client as well. Download it and install it from https://claude.ai/download

After you install it, run it and log in to your Claude account.

2. Install Node.js

We’ll have to install Node.js for the same reason we’re installing Claude for desktop. We’re running everything locally and node helps us load and run the servers.

Go to Nodejs.org and follow the instructions to do so.

3. Install your first MCP server

We’re going to start with the File System server. It’s created by Anthropic and allows Claude to access files on your computer.

The first thing you need to do is click on Claude and then Settings. Go to the Developer section and hit Edit Config.

This will then open up a folder where you’ll find a file called claude_desktop_config.json. It’s an empty file right now with a pair of curly braces {}.

Remove those braces and paste this in instead:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/<add your username here>/Documents",
        "/Users/<add your username here>/Downloads"
      ]
    }
  }
}

This configuration file tells Claude for Desktop that we have one MCP server, called “filesystem” and that it should use Node to install and run @modelcontextprotocol/server-filesystem. This server, described here, will let you access your file system in Claude for Desktop.

It also lists out the folders it can access. Be sure to add the correct path names (on Mac that is usually “Users/your-username/Documents”.)

3. Try it out!

After you save the config file, restart the Claude for Desktop application. It may take a few moments for it to start but when it does, you’ll see a little hammer icon in the bottom right corner of your chat box.

That’s our MCP tool list! We only installed on server, but it comes pre-packaged with 11 tools like creating a directory, editing files, and so on. Now you see why it’s so cool? Imagine having to build all of this yourself.

Let’s give it a test drive. I’ve given Claude access to a folder called Code where I store all my coding projects locally. I’m going to ask Claude to generate Hello World code in python and save that as a file in my Code folder.

Isn’t that cool? Now it seems simple but I can extend this to having Claude generated multiple files, organize it into different folders, and even push them to GitHub if it’s a coding project, all from my chat window.

What’s Next?

Creating the MCP was a genius move by Anthropic. They were losing the consumer race to OpenAI who are building integrations like web search. So they’ve leaned into their developer focus, having their community build MCP servers to extend Claude’s capabilities far beyond ChatGPT instantly.

Now that you understand why MCP matters for making AI truly helpful in your digital life, part 2 of this series will take you behind the scenes.

We’ll explore exactly how MCP works its magic, featuring a day-in-the-life scenario showing how different MCP servers can work together to accomplish tasks you never thought possible without programming knowledge.

Read it here and sign up below for more posts!