Model Context Protocol shifts AI from integration-heavy to intention-driven

AI agents are becoming a core part of modern business processes. But as anyone who has tried to connect an agent to real business applications knows, integration quickly becomes complex. In this blog, I explore a development that simplifies this landscape significantly: the Model Context Protocol, or MCP.

Rather than diving into bits and bytes, this blog focuses on what MCP really means in practice, why it matters, and how it changes the way AI agents interact with data and applications.

The old problem: too many tools, too much configuration

Traditionally, when you wanted an AI agent to interact with an application that didn’t have a ready-made connector, you had to build custom integrations. That usually meant working directly with APIs, defining mappings, and configuring each action in detail.

Even within familiar ecosystems, this quickly adds up. For every action an agent needs to perform, such as creating a lead, updating a contact, or retrieving order data, you often need a separate tool. In real customer scenarios, that can easily grow to dozens of tools per agent.

While this approach gives you control, it also introduces overhead. Tool creation, maintenance, permissions, and governance all take time and effort. As agent scenarios grow more complex, so does the management burden.

Enter the Model Context Protocol

The Model Context Protocol is designed to address exactly this challenge. At a high level, MCP is a universal way for large language models to communicate with data and applications.

Instead of telling an agent step by step which table to use and which fields to populate, MCP allows the agent to work at a higher level of abstraction. You instruct the agent what you want to achieve, and the MCP provides the knowledge of how that action maps to the underlying application.

The key idea is simple: move application-specific logic out of individual tools and into a shared, standardized protocol.

How MCP changes agent behaviour

With MCP, you no longer need to create a separate tool for every single action. You can add an MCP for a system, such as a CRM or ERP platform, and let the agent figure out which operations are required.

For example, instead of explicitly mapping every field needed to create a lead, you can instruct the agent to create a lead with certain information. The MCP already understands the structure of the system and translates that intent into the correct API calls.

The agent still needs clear instructions, but it gains far more autonomy in how it executes them. This leads to:

• Fewer tools to manage
• Faster agent development
• Cleaner instructions
• Easier maintenance over time

Governance and permissions still matter

One important point from the conversation is that MCP does not remove the need for control. MCP servers can enforce permissions, such as read-only access or restricted write operations. In some cases, multiple MCPs are used for the same system, each with a different scope or permission set.

This balance is crucial. Too much freedom can cause confusion or unintended actions, while well-scoped MCPs help agents stay predictable and reliable.

MCP inside and outside the Microsoft ecosystem

Within the Microsoft ecosystem, MCP adoption is moving quickly. MCPs are already available for services such as Dataverse, Outlook, Word, and PowerPoint, and they can be used directly in Copilot Studio.

Outside of Microsoft, adoption is more gradual, but the direction is clear. MCP is an open standard, and software vendors can implement their own MCP servers. This makes it possible for agents to interact with non-Microsoft applications in a consistent way.

The expectation is that serious software vendors are already exploring this. Not because it is mandatory, but because it aligns with how AI-driven solutions are evolving.

A practical example: faster innovation

During an internal hackathon, we experimented with building an MCP around a specific business scenario: return processes (RMA) in Business Central. With the MCP in place, the agent could quickly retrieve RMA statuses or even create new RMAs based on input from emails or other sources.

What stood out was the speed. Scenarios that would normally require significant setup became possible in a fraction of the time. That acceleration is one of MCP’s biggest advantages.

Is MCP always the right choice?

Despite all the benefits, MCP is not a silver bullet. Sometimes, a highly specific tool still makes sense, especially when strict control is required. MCP is about making things easier, not forcing a single technical approach.

You don’t need to wait for every application to support MCP before getting started with AI agents. MCP simply reduces friction, complexity, and maintenance as your agent landscape grows.

Final thoughts

The Model Context Protocol represents a shift toward more instruction-based, flexible, and scalable AI integrations. By reducing the number of tools and configurations needed, it allows teams to focus on value rather than plumbing.

It’s not mandatory, and it’s not magic. But it is a powerful enabler for anyone serious about building AI agents that work across systems and data sources.

As MCP adoption grows, it’s worth keeping a close eye on how it can simplify your own AI and integration strategy.