Here’s why your business should be down with MCP—that’s the Model Context Protocol from AI company Anthropic.

AI applications are getting more complicated and ambitious every day. Across industries, businesses are testing and deploying AI-based internal solutions and products that leverage proprietary data. But connecting an AI model with large amounts of your company’s data isn’t as simple as dragging and dropping a couple files into ChatGPT.

In the first few years of the AI boom, developers needed to create their own customized integrations to connect APIs from AI model providers to their data sources. This process could take engineers significant time to implement, and made it difficult for businesses to quickly swap out older AI models for newer, better options. Developers were desperate for a universal solution, a standardized process for connecting AI models with data that could be adopted across the industry. In November 2024, AI company Anthropic released their proposed solution to the AI-data integration problem: the Model Context Protocol, usually referred to as just MCP. The MCP gives AI model providers and businesses a free universal standard for connecting AI models to data sources and defining how that data (or “context,” as the AI industry refers to it) should be used. Here’s what to know about MCP. It’s a bit nerdy, but worth knowing about.

The basics of MCP It helps to begin with an analogy. “Let’s say you wanted to connect your Gmail to your Netflix account,” explains Theo Chu, a product manager on Anthropic’s MCP team. “Both of those parties would have to agree on a shared interface for sending data to each other, which can take months.” This process would need to be repeated for every new tool you want to integrate with your application. What MCP does, says Chu, is standardize how applications should “talk to” data hubs, freeing up engineers to spend more time optimizing their applications instead of spending hours developing custom integrations.

Many in the AI industry compare MCP to the advent of USB adapters. In the early days of personal computing, nearly every computer-based product had its own plugs and cords, but in 1996, a nonprofit organization called USB-IF introduced USB to the world. Gradually, USB became a universally-accepted standard, enabling one plug (of course, there are different versions of USB connectors, like Micro USB and USB-C) to connect to many ports. Like USB, MCP is being adopted by many organizations, such as Google, and Anthropic rival OpenAI. For small business owners specifically, MCP can come in handy when you want to analyze your customer or sales data. It starts with an MCP server To utilize MCP, you’ll need an MCP server. The server acts as a waypoint or bridge that connects AI models to data repositories. It also instructs the model to use specific tools and retrieve specific data. Many cloud-based data hubs, like Box, have announced support for MCP and have developed their own MCP servers, but because the protocol is open-source, pretty much any service with an API can be connected to AI models through MCP.

Plaid, the financial services company that enables applications to connect with users’ bank accounts, recently launched official support for MCP in early May. Stella Zhang, Plaid’s applied AI lead, says that a few Plaid-connected AI agents have been deployed in the last few weeks, and that “many more customers are already evaluating and building proofs of concept.” The company believes its clients will mainly use the MCP integration to develop agents capable of quickly generating business insights by analyzing Plaid’s data and diagnostics. Plaid has big plans to use MCP to help people with different roles within an organization access relevant data. For example, according to Zhang, a customer support agent could “get diagnostic information about their end customers’ Plaid-powered bank connections in real time instead of having to file a ticket with Plaid and wait for a response.” MCP is also becoming a useful tool for personal AI assistants like ChatGPT and Claude. OpenAI recently announced that members of its Enterprise and Team plans can now set up connections to external data sources even if they aren’t officially available through ChatGPT’s lineup of connectors. On X, OpenAI CEO Sam Altman wrote that “people love MCP and we are excited to add support across our products.”

Not everyone is totally down with MCP Some are skeptical that MCP will be universally adopted across the AI industry. Gopal Kuppuswamy, a cofounder of AI deployment firm Cognida.ai, says that MCP itself is fantastic—he recently helped a financial services company use MCP to inject customer intent data into a loan application assistant—but he worries that because MCP was developed by Anthropic, rather than a consortium like USB, other tech companies will still try to popularize their own protocols. For instance, in April, Google announced Agent2Agent, a protocol that the company says “complements Anthropic’s Model Context Protocol” by making it easier for AI agents to communicate with each other. Also, says Kuppuswamy, MCP is an “evolving standard,” meaning engineers will need to continuously perform maintenance in order to ensure that changes to the protocol don’t break applications. To learn more about MCP, and even how to build your own MCP server using an AI assistant like ChatGPT or Claude, check out Anthropic’s official MCP website.