Business Central MCP Server in 2026 Release Wave 2: Let AI Agents Query Your BC Data

The Business Central MCP server gets a significant upgrade in the 2026 release wave 2. A new data query tool lets AI agents build their own queries against your Business Central data, which Business Central then compiles and runs. No APIs need to be set up first.

Microsoft’s Business Central team introduced the update in a short launch video. In this article we walk through what it shows: what the new query tools do, how an agent actually uses them, the limits to be aware of, and the new controls for setting up and switching off the MCP server.

Good to know: According to Microsoft, the new MCP query tool is currently in preview and will become billable when it reaches general availability later this year.

What Is the Business Central MCP Server?

MCP stands for Model Context Protocol, an open standard that lets AI assistants connect to business systems and use their data and functions as “tools”. The Business Central MCP server makes Business Central one of those systems. An MCP host, such as Copilot Studio, Visual Studio Code or another AI client, can connect to it and work with your BC data on your behalf.

Until now, that access ran through APIs you configured in advance. You decided which APIs were available, and the agent could only get data those APIs exposed. The new query tools add a second, much more flexible route.

The New Data Query Tools

Microsoft summarizes the idea simply: the AI model in your MCP host writes the query, and Business Central compiles and runs it and returns the data. When you enable the data tool under server features in the MCP configuration, four new system tools become available to the agent:

• Find tables: search for the Business Central tables that hold the data the agent needs.
• Table schema: see which fields each table contains.
• Table relations: understand how tables connect to each other, which is what makes good joins possible.
• Data query: compile and run the AL query the agent has written, and return the results.

Importantly, what the agent can reach depends on the permissions of the user and the tables that user already has access to in Business Central. The quality of the results also depends on how well the agent writes AL queries.

How It Works in Practice: Microsoft’s Demo

In the video, an agent connected to the Business Central MCP server is asked a simple business question: get all customers, their orders and the lines on each order, and analyze what people are actually buying. Here is what the agent did on its own:

1. Found the tables it needed by searching Business Central.
2. Looked up how to write AL queries using Microsoft’s separate documentation MCP server. Microsoft notes that newer AI models are already good at this, so they often need it less.
3. Explored the table schemas and relations to see which fields exist and how the tables connect.
4. Asked clarifying questions, such as which time period and which aspects the analysis should cover.
5. Wrote and compiled the queries. When a query failed, the error messages came straight from the Business Central compiler, so the agent could often correct its own mistakes.
6. Ran the final queries and produced the analysis.

The whole run took about 30 steps and two minutes, and ended with a useful summary of customer buying patterns, including observations such as one customer accounting for almost half of the open orders. None of it needed an API or a report to be built in advance.

The Limits: The Power and Limitations of AL Queries

Because the tool is built on AL queries, it has both their strengths and their limits. As Microsoft puts it, whatever you can express in an AL query is now available to agents, but:

• Questions that cannot be expressed as a query, or only as an inefficient one, will not work well.
• Business Central limits how much data a query can return and how long it can run.

Microsoft sees the sweet spot as ad hoc analysis: questions where no existing report, page or API combines the data in the way you need.

Easier Setup, and a Switch to Turn MCP Off

The release also improves the MCP setup experience:

• A new Server features section in the MCP configuration separates server features, such as the data tool, from the APIs you expose.
• A system tools overview shows exactly which tools the enabled server features open up to agents.
• A new on/off toggle on the Copilot and agent capabilities page lets administrators activate or deactivate the MCP server completely.

NAV SEAL’s Take: Powerful, But Plan It Properly

Letting an AI agent answer data questions directly from Business Central, without building an API or report first, is a genuine step forward for finance and operations teams. A few practical points before you switch it on:

• Review permissions first. The agent can reach whatever the user can reach. If your permission sets are broader than they should be, an agent makes that much more visible.
• Start in a sandbox. The query tool is a preview, so test it on sandbox data before using it on production.
• Plan for the cost. Microsoft has said the tool will be billable at general availability, so think about who will use it and how often.
• Check the answers. An agent writing its own queries can still misunderstand a question. For figures that drive decisions, verify against a trusted report.
• Use the off switch. If your organization is not ready for agent access, the new toggle keeps the MCP server disabled until you are.

Microsoft’s landing page for the Business Central MCP server is aka.ms/bcMCP, and its integration overview is at aka.ms/bcintegration. For more on Copilot and AI in Business Central, see our Copilot & AI articles, and for another 2026 release wave 2 highlight, read about mirroring Business Central data to Microsoft Fabric.