Copilot in Dynamics 365 Business Central has demonstrated that artificial intelligence can effectively work with ERP data. It can generate product descriptions, reconcile bank accounts, and analyze sales trends. However, it operates within a closed ecosystem – inside Business Central.
Model Context Protocol (MCP) goes a step further. It is an open standard solution that allows different AI applications to access data from Business Central. A company’s custom chatbot, an assistant built in Copilot Studio, or an external AI tool – all can use the same standardized gateway to your ERP system.
Microsoft has introduced MCP support in Business Central, opening the door to building tailor-made AI solutions aligned with your company’s specific needs.
What MCP actually is
Copilot in Business Central works great for daily operations – but only inside the system. What if you want to build your own assistant, a chatbot on your company website, or connect AI to multiple systems at once?
This is where MCP assists you: a universal access gateway to Business Central. It allows you to grant different AI agents controlled access to your ERP system. Each assistant receives a “pass” with defined permissions – one may only retrieve data, another may also create or update records. You decide who and which assistant.
In practice, MCP is a communication protocol between AI applications and business systems. It was developed by Anthropic and announced as an open standard at the end of 2024. Microsoft adopted it quickly, enabling Business Central to work not only with Copilot, but with any AI tool that supports MCP.
Why MCP was developed
Before MCP, AI integrations with business systems were fragmented and chaotic. Each AI model provider (Anthropic, OpenAI, Google) built its own way of connecting to external tools. Each software vendor (Microsoft, Atlassian, Slack) had to create separate integrations for every model.
Imagine having 5 AI models and 10 business systems. Without a uniform standard solution, you need 50 separate integrations. Every update requires changes in multiple places. A solution built for one model doesn’t work with another.
MCP solves this through standardization. Instead of dozens of separate integrations:
- Each AI model supports MCP.
- Each business system exposes an MCP server.
With an MCP server, Business Central can work with any AI tool supporting the protocol, without building custom integrations for every combination.
How this looks in daily work
Copilot already lets users ask natural-language questions in Business Central: check inventory, analyze sales, or generate product descriptions. This is a major improvement over manual navigation.
MCP extends these capabilities beyond Business Central. Imagine an assistant that, with one question, checks customer data in Business Central, verifies communication history in CRM, and finds related documents in SharePoint. Or a chatbot on your website that answers customers about order status directly from your ERP.
Crucially, every MCP-connected assistant works on behalf of a specific user. It only sees what that user is authorized to see. If a user has no access to financial data, neither will the assistant, whether it’s Copilot or a custom AI tool.
What MCP adds beyond Copilot
Multi-system data integration
Copilot only sees Business Central. MCP allows building assistants or agents that query ERP, CRM, SharePoint, and other sources in a single request. Instead of switching between applications, users ask one question and get a complete answer.
Custom AI tools tailored to your business
Copilot follows Microsoft’s general design. MCP enables building custom assistants or agents in Copilot Studio that reflect your company’s processes and terminology. A sales assistant can behave differently from an accounting assistant.
Integration with external applications
A website chatbot answering customers about order status. A mobile app for field sales retrieving ERP data on the go. Automated notification systems reacting to changes in BC. MCP opens Business Central to scenarios far beyond the standard user interface.
Flexible automation
Traditional automation follows fixed rules: if X, then Y. AI agents can handle tasks that vary each time: “Review new orders and flag those requiring credit verification based on customer history and order value.” This requires contextual judgment, not simple filtering.
What it means for your business
An open standard, more possibilities
Copilot is a ready-to-use solution that works out of the box. MCP is different: an open standard enabling custom tools and integrations with any AI platform. Not a replacement – an extension. Copilot for everyday tasks, MCP for advanced, flexible solutions.
One assistant, many systems
Most companies operate multiple systems – ERP, CRM, document repositories in SharePoint, spreadsheets in Microsoft Excel. MCP allows building an assistant that connects them all. Instead of five tools, users have one place to ask questions.
Tailor-made solutions
Every company has its own processes, terminology and workflows. Copilot offers universal features suitable for most scenarios. When you need more, an assistant that understands your industry, internal procedures, and company language, MCP provides the foundation to build it.
How the MCP server in Business Central works
Model Context Protocol is based on a simple model:
An AI application (custom chatbot, Copilot Studio assistant, external tool) communicates with the MCP server built into Dynamics 365 Business Central through a standardized protocol.
When a user asks a question, the AI application recognizes that ERP data is required. It sends a request to the MCP server in Business Central. The server checks the user’s permissions, retrieves the data, and returns it to the AI application, which then formulates the response.
The key difference from Copilot: the same mechanism works with any MCP-compatible AI application. Business Central doesn’t need to know which tool it’s talking to – as long as it speaks MCP.
Security of MCP integrations
One of the first concerns for IT leaders is security. Can AI do something it shouldn’t? Are company data protected?
Microsoft designed the MCP server in Business Central with security in mind:
Read-only by default. When MCP is enabled, agent can only read data. They cannot create, modify, or delete anything unless an administrator explicitly grants those permissions.
User-based permissions.The assistant operates in the context of a specific user and has exactly the same access rights. If the user cannot see purchase prices, neither can the assistant.
Administrator-controlled configuration. Access can be precisely defined – which data agent can access and what operations they may perform. Different agents can have different permission levels.
Protection against prompt injection. Even if someone tries to manipulate the AI into performing unauthorized actions, Business Central’s permission model blocks access beyond the user’s rights. Even if someone attempts to “trick” the AI, the system will deny access to any data that the user is not authorized to view.
What MCP is not
MCP does not replace Copilot. They serve different purposes. Copilot is a ready-made assistant. MCP is infrastructure for building your own. Many companies will use both.
MCP does not replace the ERP system. Data remains stored and processed in Business Central. MCP is simply an additional access layer.
MCP does not replace traditional process automation. Power Automate handles deterministic workflows: the user defines rules (“if X, then Y”), and the system executes them in exactly the same way every time. MCP with AI agents enables cognitive automation: you define the goal, and the agent determines how to achieve it.
When to use it? Power Automate is well suited to repetitive, predictable processes such as invoice approvals, notification delivery, or data synchronization. MCP and AI agents are better aligned with unstructured tasks, such as “analyze why sales dropped in this region” or “identify customers who may be interested in a new product.” Many organizations will use both tools in parallel, each applied where it delivers the greatest value.
And MCP is not a magic solution. AI assistants and AI agents can make mistakes, misunderstand questions, or miss relevant data. It’s a powerful tool but one that requires thoughtful design and adaptation to business needs.
Status and availability
The MCP server in Dynamics 365 Business Central is available from version 27.1 (2025 Release Wave 2) as a Public Preview. This means the feature is ready for testing, though Microsoft may still introduce changes before general release.
To start using MCP integration, you need:
- Dynamics 365 Business Central version 27.1 or later
- MCP enabled in system settings
- Microsoft Copilot Studio to build and manage assistants
Where to start
MCP is a new technology that requires a strategic approach. Before building AI agents, it’s worth answering:
- Which business processes could benefit from AI integration?
- Which data should agent access and which should remain restricted?
- How does MCP fit into your broader automation strategy?
As a Microsoft partner, we help companies assess MCP opportunities and plan their first implementation steps.
Summary
Copilot in Dynamics 365 Business Central has demonstrated that artificial intelligence can effectively work with ERP data. Model Context Protocol goes further, opening the ERP system to the entire AI ecosystem and enabling custom integrations.
For businesses, this means the ability to build proprietary agents, connect data from multiple systems, and create solutions tailored to internal processes. Not instead of Copilot but alongside it, as an additional layer of flexibility.
The MCP server is now available in preview. This is an excellent moment to explore whether custom AI tools can streamline work in your organization.
