What the heck is Model Context Protocol (MCP)? And why is everybody talking about it?

If you haven't been living under a rock the last few weeks, you've probably seen people talk about MCP (aka Model Context Protocol). It's popping up everywhere: on X, LinkedIn, Bluesky, in dev forums on Reddit, and even in random coffee chats.

But what is it, exactly? And why is it suddenly the thing that everyone's talking about? 🤔

Let’s start at the very beginning. MCP is short for the Model Context Protocol, an open standard released by Anthropic in late 2024.

You can think of MCP as the USB-C of AI integrations that lets the Large Language Models powering tools like Claude or ChatGPT communicate with external data sources and tools (Obsidian, Gmail, Google Calendar, and even my favorite The Cat API 😼) without needing a million custom integrations.

Source: Noharsakal

It's essentially giving your AI application access to whatever context you need it to have; the ability to peek outside its dataset and grab relevant info or take actions in your digital world. And because MCP is an open standard (Yay! Thanks Anthropic 💜), anyone can build on it, which is one of the reasons why the tech community's losing their minds over it.

So, what does MCP actually do?

Hopefully, now, you get the gist of MCP (if not don’t worry I have loads of examples for ya). So.. what does MCP actually do?

Imagine an (AI) assistant that only has access to old data, is clueless about your files, and is only good for chatting. (Unless you use Pieces LTM, then you never have to deal with this 😉)

MCP fixes that by connecting your AI to the stuff you care about, your database, your GitHub repo, your calendar, etc.

How does it do that? Well, MCP is an open standard from Anthropic that lets LLMs securely access the context they need and take action on it. 

It uses hosts, which are AI apps like Claude Desktop or Cursor that you would like to connect to the outside data.

Each host has clients, they are like connectors that pair up 1:1 with servers. Those servers are small programs that link to specific tools or data, like your local files or a database on your machine, or tools/data sources like GitHub or a cat API.

For example, recently Jim built his own MCP server that Claude’s client can call to tap into Pieces LTM-2 on your computer. 

The client talks to the server through MCP, and voilà, your LLM can access all the info or even trigger actions on your behalf. 

It works by naming tools and resources, so when you write a prompt, the LLM can figure out (non-deterministically) whether it should call a tool or use some data. 

Everything runs through the AI app you’re already using, like Claude or Cursor, and the MCP layer handles the coordination behind the scenes. It’s secure, scalable, and (most importantly) saves you from writing endless custom code.

Which tools and platforms support MCP?

More and more tools are starting to support the Model Context Protocol (MCP), and honestly, it’s been pretty cool to see how it’s showing up all over (some of my recent favorite releases are by Microsoft!). These MCP integrations make it easier for tools to access context and memory so your assistant (or copilot) actually knows what’s going on. To me, this all feels like part of a broader shift toward better communication between tools, long-term memory across workflows, and giving agents the context they need to take meaningful action.

MCP Clients: tools that use the Model Context Protocol

One of the biggest signs that the MCP hype isn’t going to die down anytime soon is the growing number of tools that already support it (and the ones adding support every day). These tools act as clients. They use MCP to orchestrate agents, call tools, and pass along real-time context.

Check out some of these popular MCP clients:

• Microsoft Copilot Studio: In March 2025, Microsoft added support for MCP to Copilot Studio. This means that the agents you build can now connect to external tools and data sources through MCP.

• Claude Desktop: As part of Anthropic’s implementation of MCP, Claude’s desktop app supports MCP, which means it can hook into external tools to control agents or return data. It’s a step toward making Claude feel more like an assistant that understands your entire workflow rather than just a chat window.

• Cursor: This code editor integrates MCP to connect directly to external systems like databases, Notion, GitHub, or Stripe. Instead of manually feeding in project structure or data, Cursor can just ask for what it needs.

• Github Copilot: Agent mode integrates MCP to connect with external tools and services. It can pull in context, call actions, and use real data to help complete coding tasks for you.

• Zed: Zed’s adoption of MCP lets this fast, collaborative code editor connect to external context servers like Postgres or even custom ones you build yourself. That way, it can bring in the info your AI assistant actually needs to help you out while you’re coding.

• Block (formerly Square): Codename Goose is Block’s open-source AI agent that uses MCP-powered extensions to automate tasks like migrating code, setting up monitors, or connecting to tools like GitHub and JetBrains.

• OpenAI: : The OpenAI Agents SDK now supports MCP, which means agents can connect to external tools and data sources through any compatible MCP server. Instead of hardcoding access to APIs or services, you can just hook them up through MCP and let the agent call what it needs. 

All of this just shows how MCP is quickly becoming part of the infrastructure allowing all of our favorite tools to have the context they need.

MCP servers that developers can use

MCP servers are the other half of the equation. These platforms expose a Model Context Protocol-compatible API so tools like Cursor or Claude can connect, fetch data, and perform real actions.

Here are a few MCP servers that you should check out:

• Custom MCP Servers: MCP is flexible. You can actually build your own MCP server in any language that supports stdout or HTTP. Microsoft and Anthropic actually JUST released (April 2nd) an official C# SDK for MCP, making it easier to create custom servers and integrations in .NET. Also, make sure to check out some of the other SDKs that you can build with: Python SDK, TypeScript SDK,  Kotlin SDK, Java SDK, Rust SDK, and Swift SDK.

• Appwrite: Appwrite offers a built-in MCP server that lets LLMs interact directly with its APIs. Tools like Claude Desktop and Cursor can plug right into your Appwrite project without any extra setup.

• Playwright MCP Server: An MCP server from Microsoft that lets LLMs control browsers using Playwright. Clients can run end-to-end tests, click buttons, or fill forms by sending instructions in MCP format.

• Baidu Map MCP Server: Baidu Map is the first map service provider in China to support the MCP protocol. It offers core APIs for geocoding, route planning, and weather queries, all accessible to MCP-compatible assistants through a server-side API key.

• Perplexity Ask MCP Server: An MCP-compatible server that connects to the Perplexity (Sonar) API to enable real-time web search. It lets tools like Claude fetch live results directly through the MCP ecosystem which is especially useful for research or answering questions that need access to recent info.

There are tons of other MCP servers out there. Everything from search to storage to stuff you probably didn’t even know you needed. If you’re looking for something specific, check out mcp.so or mcpmarket.com to see what’s out there.

How can developers start using Model Context Protocol?

Let’s ditch the theory and get to the good stuff: what devs are actually doing with MCP. Don’t worry, we’ll show you how to do it yourself in the next article of this MCP series (Spoiler: it may or may not involve cats).

But first, check out some of these interesting use cases and personal projects that people have been sharing all over socials.

Bring context to your workflow

Imagine you're in Cursor, and you hit a bug. Normally, you'd dig through logs or Stack Overflow yourself (or hey you might just use Pieces 🤷‍♀️). 

But with Model Context Protocol, Cursor can connect to an MCP server tied to your project's database or error logs. You can directly ask, "Why's this crashing?" and the AI pulls the latest error data, reads your codebase, and suggests a fix in seconds. @PrajwalTomar_ called this combo "10x faster" for debugging, and it's easy to see why.

Accessing Pieces Long-Term Memory in Claude

Jim Bennett shared a super interesting implementation on LinkedIn: "Want to use Pieces for Developers Long-Term Memory in Claude instead of inside Pieces? Well, you can using MCP – model context protocol. The Pieces C# SDK makes it a low lift to write an MCP server that can be called by Claude that can interact with Pieces." 

He even published the demo code on GitHub, showing how easily Claude can utilize Pieces Long-Term Memory Agent through an MCP connection. This integration is a great example of how MCP can connect different AI tools. I highly suggest you fork Jim’s repo and try it out for yourself! 

Managing your day

Here’s a cool example from @cline: you’re using Cline as your personal assistant. You ask, "Reschedule my 2 PM meeting." Without MCP, Cline has absolutely 0 info about your calendar. With MCP, it connects with a server linked to Google Calendar, checks your available time slots, moves the meeting, and even can send an email to the attendees. No more switching apps. Just one command and you’re done. 

Automating stock market data

Check this out from @virattt: he built an MCP server connecting Claude Desktop to real-time stock market data via Financial Datasets AI. You can literally ask 'What's the current price of AAPL?' and it flows through the MCP host, client, and server to send the real-time price back to you.

What’s the future of MCP?

MCP’s been around since November 2024 (I know, it feels like it just popped off the last few weeks). So why’s it suddenly gone mega-viral now?

Blame the snowball effect: @alexalbert__’s launch post racked up 1.4M views on X, devs like @angiejones.tech shared an amazing guide on Agentic AI and the MCP ecosystem, and @mattpocock.com released a full tutorial series. Plus, tools like Claude Desktop and Cursor joined in supporting MCP – and most recently (March 19, 2025), Microsoft announced MCP in Copilot Studio. 

It’s hit a sweet spot where the hype is meeting real results, and the fact that it’s open standard makes it even better.

So, what’s next? 

I genuinely think MCP’s going to change how we use AI. You could build your own assistant that scans your GitHub commits to catch bugs before you do, or one that reads your calendar and reminds your team about a deadline. 

You could even have it pull live stock prices and warn you to sell (or buy).

We’re only at the beginning, but it’s already inspired a wave of amazing new AI projects, and I can’t wait to see what people end up building next.

That said, if you’re interested in learning more about MCP, check out the official MCP docs and GitHub repo – or just wait for my next article, where I’ll show you how to build your own project with MCP 😉.