What is the Model Context Protocol (MCP)?

What the Model Context Protocol is, how it works, and why it’s being called "the USB-C for AI tools"

The Model Context Protocol (MCP) was introduced by Anthropic in November 2024. On the official site, it’s described as:

“An open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools.”

But what does that actually mean in real life?

In practice, MCP is a way to connect tools and applications to your language model (e.g. to Claude, or ChatGPT) — so that the model can pull in data (‘context’) or take action on your behalf.

Think:

• Searching your WhatsApp Messages

• Summarizing a Slack thread

• Pulling data from your local files

• Querying or analyzing data from your database

With MCP, all of that can happen without switching apps. The model can access and act on that context inside the chat window, using natural language.

A more familiar example

If you’ve used an AI chatbot with a “Search the web” feature, you’ve already seen a version of this idea: The model doesn't just rely on its training data — it reaches out to a live tool (in this case, a web search) to get better answers.

MCP expands that idea, letting developers plug in a range tools — via something called MCP servers.

Here’s an example of what that looks like in Claude Desktop:

Example: Searching WhatsApp from your LLM

An early viral implementation of the Model Context Protocol was for connecting WhatsApp to an LLM (e.g. Claude).

Once it’s set up, you get new WhatsApp-specific tools available directly in your model interface, like:

• get_chat — find a specific WhatsApp chat

• get_direct_chat_by_contact — pull your DMs with a specific number

Here’s what that looks like when you click on the ‘🛠️’ within Claude Desktop:

I don’t have to explicitly tell Claude to use the WhatsApp MCP — it just knows those tools are available, and calls them when they’re relevant.

So now, I can ask:

“Can you search my messages to find the most recent AI Club events?”

And Claude can fetch that context directly from WhatsApp:

The value of a single standard

While this is a fun and specific use case, the real power of MCP is that it’s a universal standard.

Many people — led by Anthropic — have started calling it “the USB-C for AI applications:”

Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.

As someone whose AirPods still use the old charger, I mildly resent the USB-C analogy. But I do think this visual from Norah Sakal really helps show why the USB-C metaphor is worthwhile:

Norah Sakal’s amazing drawing of how MCP acts as a standardized connector.

Before & After MCP

Before MCP:

• Connecting one tool (like Slack) to one AI (like Claude) meant building a custom integration

• That doesn’t scale in a world with lots of tools and lots of models

With MCP:

• Tools become MCP servers

• Models (or apps like Claude, Cursor, or Windsurf) act as MCP clients

• You can mix and match freely

In short: MCP solves the many-to-many integration problem with a single universal standard.

What is the status of MCP today?

Since it was introduced in late 2024, interest in and adoption of the Model Context Protocol has been growing rapidly:

• Anthropic's Claude Desktop has supported MCP from early on

• AI developer tools like Cursor and Windsurf quickly followed

• Open-source communities have built hundreds of MCP servers

• And new tooling is helping developers build, host, and manage MCP integrations more easily

Most recently, on March 26, OpenAI also expressed their intent to support MCP:

Tweet Source

And today, after polling his Twitter/X followers for input, Sundar Pichai announced that Google would hop on the MCP train, too:

Tweet Source

(Though…. Google also launched a ‘complementary’ agent-to-agent standard called A2A. The jury is still out on whether A2A and MCP will be truly complementary vs. competing).

Where are we headed?

The most exciting part is that we’re still very, very early.

In future posts, I’ll dive into:

• How MCP actually works (spoiler: JSON, RPC, and a few terminal commands — for now)

• Challenges and open questions

• Shifting from ‘local’ to ‘remote’ MCP

• Open questions around billing/monetization, security, logging, and more.

Thanks for reading — and if you’re building something in this space or have an MCP-related topic you want to know more about, I’d love to hear from you.