MCP: A Simple Guide To Going Further With AI

If you've been following AI developments lately, you've probably noticed a new acronym taking the industry by storm: "MCP."

The Model Context Protocol might sound like technical jargon, but it represents a significant shift in how AI applications will connect with real-world tools and agents.

Over the past few weeks, I've been testing MCP extensively with various systems, and I'm very excited about what this means for builders, businesses, and users alike.

Today, I’m breaking down:

• What MCP actually is (in plain English)

• Why it fundamentally changes what's possible with AI

• How HubSpot just became the first major CRM to support it

—@dharmesh

What is MCP?

MCP (Model Context Protocol) is an open standard that's exploding in popularity among early adopters in the AI industry.

In simple terms, MCP is a protocol that allows AI applications (the "MCP Clients") to communicate with various services (the "MCP Servers"). Think of these servers as offering specialized capabilities—or "tools"—that the AI can leverage when needed.

What makes this revolutionary is the standardization. With MCP, any AI that speaks the protocol can instantly tap into an (exponentially growing) ecosystem of servers. This dramatically amplifies what systems like Claude, ChatGPT, and agent.ai can accomplish without custom coding for each integration.

It's kind of like what HTTP did for the web or what USB did for hardware connectivity.

Here’s an example:

I've configured Claude Desktop to connect with multiple MCP Servers from various companies. Now when I type a prompt, the LLM can intelligently choose from hundreds of tools based on my request:

• Use agents on Agent.ai

• Access data directly in my HubSpot CRM

• Manage files in specific directories on my computer

• Post updates or retrieve information from Slack

• Interact with my Google Calendar and Gmail accounts

Just the other day I used the prompt: "Find OpenAI in my HubSpot CRM and send me the details on Slack, including when we last connected."

With one natural language request, the AI navigated multiple systems to deliver exactly what I needed. And that's just an overly simple example to illustrate what's possible.

Why does MCP matter?

LLMs have historically been limited to the data they were trained on. This is what's often known as the "knowledge cutoff date."

To work around that limitation, two major innovations emerged:

1. RAG (Retrieval-Augmented Generation): whereby we put relevant information right into the context window. (We retrieve that relevant information by using what’s called a vector database—but that’s a topic for another day).

2. Tool Use: whereby the LLM is given access to a set of "tools" it can use to respond to a prompt

The problem with tool use has been that for a tool to be available in the LLM, it had to be specifically coded into the AI application (like ChatGPT or Claude). Adding new tools required updating the AI application itself.

MCP simplifies and decouples things. Think of it like the USB-C standard.

If you build a new laptop, all you need to do is support the USB-C standard and add USB-C ports. Then all existing USB-C devices will work with your laptop. Similarly, if you make a new peripheral device that supports USB-C, it will work with all computers that support the standard.

That's exactly what MCP does for AI applications:

• AI applications (Clients) can connect to any MCP Server

• Applications and tools (Servers) only need to implement the MCP standard once

Before MCP, having your CRM data accessible to three different AI systems required three separate integrations. Each developer would need to know how to use the API (Application Programming Interface) of the CRM and then write specific code for each possible use case, calling the API as needed.

Now, implement the standard model context protocol once, and you're compatible with every AI app, agent or system that speaks MCP.

The real magic happens when you combine tools from multiple MCP servers to accomplish higher-level goals. I predict by the end of the year, there will be thousands of MCP Servers that will make it possible to power AI apps and agents with all sorts of data and capabilities.

HubSpot's MCP Launch: A Perfect Example

"Find all the deals in my pipeline that haven't moved in 30 days, analyze the last conversation notes for each, and suggest personalized next steps based on the customer's history."

Imagine typing that into your AI assistant and getting a complete analysis and action plan in seconds. That's what's now possible with HubSpot's new MCP integration.

I announced it yesterday on LinkedIn (very excited about it) but HubSpot just became the first major CRM platform to launch MCP support, making the data and functionality of our platform accessible to any MCP-compatible AI application.

That means the 225,000+ companies using HubSpot to manage their marketing, sales, and customer service can now leverage AI in ways that were previously impossible without extensive custom development.

This new feature allows HubSpot to be accessed easily by the growing list of AI applications and agent platforms that support MCP. MCP adoption is on fire (🔥), and I'm really thrilled that HubSpot is all over it.

The implementation is simple. You first go to HubSpot and after proper authentication, you get an access token and an MCP server link. You go to Claude (and soon ChatGPT) and configure it to "point" to the HubSpot MCP server. Claude will then query the server and discover 21 powerful tools available (most based on existing APIs) —everything from looking up records to creating contacts to associating companies with deals.

Here's the beauty of it: Once the HubSpot MCP server is made available, the AI application (acting as the MCP client) gets superpowers.

You can write prompts like:

"Find all contacts added in the last 30 days who are based in Boston, create a list of them, and draft a personalized email about our upcoming event there."

Or:

"Analyze our sales pipeline, identify deals that haven't moved in 30 days, and suggest next steps for each based on the last conversation notes."

The real magic is when you combine tools from multiple MCP servers to accomplish higher-level goals.

One simple prompt. Multiple systems. Real business value.

The Future of AI is Connected

I've been working in tech for 30+ years, and MCP is one of those rare innovations that deserves every bit of the hype it's getting.

It’s funny how many people get intimidated by the technical jargon. "Model Context Protocol" sounds like something from a sci-fi movie, but in reality, it's a simple idea: let AI talk to your favorite tools using a standard language they all understand.

That's it! No master’s degree in machine learning required.

HubSpot's early adoption of MCP is just the beginning. By year's end, I expect thousands of MCP servers across every industry, creating an explosion of possibilities for what AI can actually accomplish for businesses.

The companies that move quickly here will have a big advantage. Their users will be able to leverage AI in ways that create genuine business value, not just viral demos or theoretical capabilities.

So, are you using any MCP servers yet? I'd love to know what you're experimenting with! Vote in the poll below or just reply to this email and let me know.

—Dharmesh (@dharmesh)