Here’s how to seamlessly connect an AI model with your company’s data

We’re almost at the point of being blasé about technological advancements that would be described as revolutionary only 12 months ago. However, in November 2024, Anthropic released a concept which, although it flew under the radar for a little while, has the potential to completely revolutionize the relationship between AI and the data you own.

Data gap

Why do I think it’s the best thing to happen to AI in 2025? While it’s clear that the widespread adoption of generative AI, GPTs, and agentic tools benefits business and individuals, boosting productivity, workflows, and automations, there has always been a gap between the model and the data we want to interact with.

This data gap has been previously filled through expensive model re-training and custom tool development, but limited to those with big budgets or technical knowledge. Model context protocol (MCP) provides an answer to this.

MCP sits as a server running between your LLM and other systems, acting as an interpreter of complex and disorganised data by providing context to the AI.

Making data available to external systems isn’t new; microservices and APIs have been around for decades. But they require a human to have access to the documentation to understand the capabilities of the API, and once the integration is in place, it provides only a snapshot of functionality. APIs change over time, and the software using them needs to be updated.

This is why MCP is so exciting for technical and non-technical users. Suddenly, we can now implement an interface between disparate systems that continually keeps pace with changes in their capabilities with little to no overhead in effort or cost.

Unifying content

Consider the number of systems running that contain your business data. Written documentation, customer relationship management software (CRMs), order management systems (OMSs), databases, and APIs. Some of these might be customer-facing as well as tools used by internal teams.

With MCP, the content of these silos can be unified and made accessible through an AI client like Claude desktop to provide a more personalized service to customers, or to help staff access more in-depth information far more easily than before.

This can allow for easy ordering. In a retail environment, MCP could connect to an OMS to streamline order processing. Chatbots could access product availability through real-time data from the OMS and allow customers to buy products directly through a chat interface.

It could also help provide a personalized customer-support experience. By integrating with a CRM, an LLM could access customer purchase history, preferences, and previous interactions via MCP. This would allow it to offer tailored recommendations and resolve issues more efficiently. This kind of use could enhance the customer service experience and increase customer loyalty.

Though exciting, at the moment, it’s only possible to describe MCP as a promising open-source standard.

Future adoption

One of the main limitations of MCP is the limited number of clients that currently support it. While MCP has the potential to improve how external data and tools connect with AI models, its adoption is still in the early stages.

Additionally, only a few providers offer MCP servers, which restricts its availability and scalability. This limited support can hinder the widespread implementation of MCP and its benefits.

In terms of future adoption, while the principle behind MCP has been designed by Anthropic, its success depends on adoption by other AI providers. In March 2025, OpenAI announced that it would adopt MCP as a standard and is expected to make it available to its clients soon.

This adoption by major AI providers like OpenAI is crucial for MCP to become a universally available standard, and as more AI providers embrace MCP, the number of integrations, its capabilities, and benefits will become more accessible to a broader range of applications and industries.

Being backed by Anthropic and OpenAI, it seems inevitable that MCP will become the chosen standard for AI clients. As a result, more systems will provide official MCP servers that handle interactions between the two.

The current stable of MCP servers is targeted at early adopters, focused on tech teams and developers, but if the standard progresses at the same pace as innovation across the rest of the AI sphere, it won’t be long before plugging your Copilot agents into your sales pipeline is as natural and quick as asking ChatGPT to write your next LinkedIn post.