MCP Server in Business Central: opening ERP data for AI agents

Key takeaways

• The MCP Server in Business Central enables external AI tools — Copilot Studio, Claude Desktop, custom agents — to access ERP data in a structured way.
• Setup is done via Feature Management and MCP Server Configurations, where API pages are published as tools.
• Business Central's security model applies unchanged: the AI agent has exactly the permissions of the assigned user.
• MCP replaces individual connectors with an open standard, making integrations more robust and easier to maintain.
• Typical use cases: AI-powered data queries, automated workflows, internal chatbots with ERP connectivity.

Why MCP instead of traditional interfaces?

Traditional integrations between AI tools and ERP systems typically require individually developed connectors: API calls are programmed manually, data structures are hardwired, authentication is handled separately. Every change to the ERP or AI tool requires adjusting the connector.

MCP solves this through standardisation. The protocol defines how an AI agent discovers available tools, what data they provide and which actions are permitted. The agent does not need to know Business Central's internal structure — it asks the MCP Server what is possible and uses the tools offered.

This significantly reduces development effort and makes integrations more robust: when the API structure in Business Central changes, the tool adapts without the agent needing reprogramming.

Setting up the MCP Server in Business Central

Setup takes just a few steps directly in Business Central Online.

• Activate the feature: navigate to Feature Management and enable the 'Enable MCP Server access' option.
• Create a configuration: search for 'Model Context Protocol (MCP) Server Configurations' and create a new configuration.
• Define tools: add the desired API page objects to the available tools list — standard APIs or your own custom API pages.
• Set permissions: for each tool define whether the AI agent may read, create, modify or delete.
• Connect AI host: transfer the connection details to your AI host — whether Copilot Studio, Claude Desktop or a CLI-based agent.

Security model: no bypass, no special rules

A common concern when connecting AI tools to ERP data is access control. With the MCP Server in Business Central: the agent has exactly the same permissions as the user it runs under.

There is no bypass of the BC permission model. No separate security concept that needs additional maintenance. No access to data the agent's permission set does not allow.

In practice this means: you create a dedicated user or app registration for the AI agent, assign the appropriate permission sets and define via the MCP configuration which tools are available. The combination of BC permissions and tool authorization gives you granular control.

Practical use cases

The MCP Server opens up a range of possibilities that go beyond the built-in Copilot features.

• AI-powered data queries: a sales representative asks in Claude Desktop 'Which customers haven't placed an order in the last 90 days?' — and gets the answer directly from BC data.
• Automated workflows: in Copilot Studio you create a workflow that checks open items daily and sends a notification to sales when payment deadlines are exceeded.
• CI/CD pipelines: developers can query BC data in their deployment pipelines — for example, to verify after an extension installation that certain records were created correctly.
• Internal AI assistants: you build an internal chatbot that answers employee questions about inventory levels, delivery status or invoice status — without them having to log in to Business Central.

Our assessment: when MCP makes sense

The MCP Server is worthwhile when you want to connect AI tools to your ERP data beyond the built-in Copilot features. Typical scenarios are: custom AI assistants for employees, integration with external AI platforms such as Copilot Studio or Claude, and automated workflows that access BC data.

MCP is less useful when your requirements are fully covered by the built-in Copilot features and the standard agents (Sales Order Agent, Payables Agent). In that case you don't need an additional integration channel.

Frequently asked questions

Do I need programming skills to set up the MCP Server?

The basic setup — activating the feature, creating a configuration, publishing standard APIs as tools — requires no programming. If you want to provide your own custom API pages as tools, you need AL development skills.

Is the MCP Server available on-premises?

The MCP Server is a feature of Business Central Online. On-premises installations cannot use MCP.

Which AI hosts are supported?

The MCP Server implements the open Model Context Protocol standard. Any MCP-compatible host can connect — including Microsoft Copilot Studio, Claude Desktop, CLI-based agents and custom applications.

Can AI agents modify data via MCP or only read?

Both are possible. In the MCP configuration you define per tool whether the agent may read, create, modify or delete. Business Central permissions apply as an additional safety net.

What is the difference between MCP and the built-in Copilot features?

The built-in Copilot features (chat, agents) work within the Business Central interface. The MCP Server, on the other hand, opens BC data to external AI tools — for use cases that go beyond the standard interface.