Hi Tip Sheeters! This week I will take my first look at Model Context Protocol (MCP), which Anthropic rolled out during the holidays last year. Along the way, I'll share my approach to getting up to speed on a new open-source framework or tool. Let's get started!

7 Steps to get up to speed on MCP (or any new tech)

There are some repeatable steps about how I get to know a new tool. One thing to remember up front is that just because someone proposes a standard or releases a tool doesn't mean it will become important or influential. But many will become useful and interesting.

Step 1: Find the trending tools

First you have to be looking for new tools and tech. Sources change over time, but in the last few years, I have mostly used LinkedIn and newsletters for data science related news. Other people have good luck with TechCrunch, Medium, and Reddit.

MCP was announced last December, and the number of people referencing it on LinkedIn has been growing recently such as these:

Step 2: Decide if a new tool or technology matters (to you)

You can't keep up with it all -- big surprise, right? So you need some kind of filter or frame to view the tech news through. I don't have a literal filter on my sources, but as I read headlines I know what I'm looking for. I'm looking for API-related technologies and stories in AI and data science.

MCP matters to me because it's about connecting data sources to LLM applications. It's a way of solving the knowledge gap, where an LLM doesn't know specific detailed data. One of the key data sources you'll want to connect an LLM to is your API.

Step 3: Find the official announcement

Often, a major new technology is created or sponsored by an influential technology company. They generally post an initial blog post explaining the purpose and giving some initial details. MCP was announced in November by Anthropic, a U.S. LLM maker known for the Claude models. The initial blog post was titled Introducing the Model Context Protocol.

That post explains the problem that MCP is trying to solve:

Then it gives the official definition: "The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools. " It also shares some initial resources in a Getting Started section:

Step 4: Check out the project page

Even when created by a commercial company, new technologies are often officially open-sourced. They usually have a project page separate from the company's website, along with a GitHub repo with any related code.

For MCP, the project page is https://modelcontextprotocol.io/. This page includes has a basic diagram that gives a high level of the components:

For my purposes, the key is at the bottom of the diagram, showing that to create a component called an MCP Server and point it to our APIs. One thing to notice here -- even though the software component is called a server, this doesn't necessarily mean an EC2 instance or separate computer - the server is software that we write and run.

Step 5: View some independent blogs or videos

Outside of the official sources, you can track down other developers or data scientists who research the topic and post explainer articles or videos. Here are a few good ones that I've found for MCP:

​What is MCP? Integrate AI Agents with Databases & APIs​

​Model Context Protocol: Claude's NEW feature explained in 2 minutes​

Step 6: Put it in context with other key tools

Here you want to figure out how this either competes with or works with an existing tool or framework. Just because a company proposes a new tool doesn't mean the market will adopt it. There are usually competing approaches.

For MCP, I'm still trying to figure this context out a bit. I think the most direct competitor is OpenAI's approach to connecting applications to their models. Since their models got immediate traction in the marketplace, some of the other LLM makers used this format or had a kind of wrapper for it. What's less clear to me is if MCP is also a competitor to the Agent frameworks like LangGraph and Autogen, or if they'll work together. I need to do some more digging on this question.

Step 7: Try it out for yourself

Once you've built a foundation with the first 6 steps, you're ready to run some sample code. It helps if you have a base project that you routinely use and know the ins and outs. (I have a base fantasy football API that I created for my book, and I can use it in a variety of ways.)

I like to use GitHub Codespaces for these sample projects, because you can install a bunch of random open-source libraries and work in a virtual environment. When you're done you can shut down the Codespace and you don't have to worry about it interacting with any other projects you have going.

I'll be working on Step 7 for a future Tip Sheet, so keep an eye out for that one. In the meantime, if you take MCP for a spin, please email me your work, I'm interested in learning from you.

That's all for this week!

Keep coding,

Ryan Day

​https://handsonapibook.com/​