Data query tools in the Business Central MCP server: a hands-on test guide

Data Query Tools (Preview) is a server feature on the Business Central MCP server. Switch it on and an agent gets four system tools: it can search your tables, read their fields and relations, write an AL query, have Business Central compile it, and run it.

What makes it useful is what it removes. Until now an agent could only read what an admin had published as an API page or API query, so anything else meant a development request and a deployment. With the data query tools, an agent can answer a question nobody built an object for, which describes most ad hoc analysis. The boundaries become the permissions the signed-in user already has and what an AL query can express. It is read-only, so nothing can be created, changed or deleted through it.

This guide takes you from nothing to a working. 

Before you start

• A Business Central online sandbox on a version that has the Data Query Tools server feature. It shipped as a preview in 2026 release wave 2.
• The MCP - ADMIN permission set, or equivalent, to create MCP configurations.
• Visual Studio Code with the GitHub Copilot extension.
• Your Microsoft Entra tenant ID, environment name, and company name.

Step 1: create an MCP server configuration

1.      In Business Central, search for and open Model Context Protocol (MCP) Server Configurations. The direct page is 8351.

2.      Select New.

3.      Set Name to something you will recognise in telemetry later. DataQueryTest works, and you will see that exact string in the configurationName dimension in step 9.

4.      Set Description to whatever you like.

Leave Unblock Edit Tools off. You are testing read-only analysis, and there is no reason to hand write access to an agent while you do it.

Step 2: turn on the data query tools

1.      Find the Server Features section on the configuration.

2.      Enable Data Query Tools (Preview).

3.      Look at the system tools box next to it. It lists the tools this feature unlocks. If you enable more than one server feature, the box shows the combined set.

You should see four tools appear:

Tool	What it does
bc_data_find_tables	Finds tables by name or semantic similarity, returns table IDs and names
bc_data_get_table_schema	Returns the fields on a table
bc_data_get_table_relations	Returns incoming or outgoing relations for a table
bc_data_query	Compiles and validates an AL query. Returns schema only unless returnData is true

That is the whole configuration for this test. You do not need to add a single API page. That is the point of the feature.

 Turn Active on.

Step 3: copy the connection string

1.      With the configuration open, select Advanced then Connection String.

2.      Copy the JSON, or select Download to save it as a text file.

It looks like this:

"businesscentral": {

   "url": "https://mcp.businesscentral.dynamics.com",

   "type": "http",

   "headers": {

      "TenantId": "aaaabbbb-0000-cccc-1111-dddd2222eeee",

      "EnvironmentName": "Sandbox",

      "Company": "CRONUS USA, Inc.",

      "ConfigurationName": "DataQueryTest"

   }

}

Step 4: add the server to VS Code

1.      Open VS Code on a folder you can keep. You will be saving .al files here later.

2.      Create .vscode/mcp.json if it does not exist.

3.      Paste the connection string inside the servers element:

{

  "servers": {

    "businesscentral": {

      "url": "https://mcp.businesscentral.dynamics.com",

      "type": "http",

      "headers": {

        "TenantId": "aaaabbbb-0000-cccc-1111-dddd2222eeee",

        "EnvironmentName": "Sandbox",

        "Company": "CRONUS USA, Inc.",

        "ConfigurationName": "DataQueryTest"

      }

    }

  }

}

4.      Save. A small toolbar appears above the server entry in the file. Select Start.

5.      Sign in when prompted. Authentication is OAuth 2.0 against Microsoft Entra ID, and VS Code uses a preregistered application, so there is nothing to register yourself.

6.      The toolbar text changes to Running.

If you are testing from Claude, ChatGPT, or GitHub Copilot CLI instead, you point at the same URL but you have to register your own Entra application first. The Learn article on non-Microsoft hosts covers that.

What a healthy start looks like

Open the Output panel and pick the channel for your MCP server. A good run reads like this:

[info] Starting server businesscentral

[info] Connection state: Running

[info] Discovered 4 tools

The line that actually matters is the last one. Discovered 4 tools means the data query tools came through.

Step 5: confirm the four tools are really there

This is the step people skip, and then they spend twenty minutes wondering why the agent keeps reaching for an API tool.

• Open GitHub Copilot Chat in Agent mode with Ctrl+Shift+I.
• Type # in the chat input. That lists every available tool by name, so typing #bc_data filters straight to the four you care about.

The mcp.json code lens is the quickest sanity check of all. Once the server is running, it shows the tool count above the server entry, which is where the Discovered 4 tools figure from step 4 comes from.

Step 6: smoke test, one tool at a time

Run these three prompts before anything ambitious. Each one is designed to force a single tool, so when something breaks you know exactly what broke.

Prompt 1, discovery:

Using only the Business Central data tools, find tables related to item ledger entries. List the table IDs and names you get back. Do not write a query yet.

Expect a call to bc_data_find_tables and a list including table 32, Item Ledger Entry. If you get an apology about not having access to Business Central, the server is not running or the tools are not available.

Prompt 2, schema:

Get the schema for the Value Entry table and show me every field whose name contains "Amount".

Expect bc_data_get_table_schema and a field list including Sales Amount (Actual) and Cost Amount (Actual). This also tells you whether the tool returns data types, which the model needs, because aggregates other than Count only work on Decimal, Integer, BigInteger or Duration fields.

Prompt 3, compile without running:

Write an AL query that returns the No. and Name of the Customer table, compile it with the data query tool, but do not return any rows.

Expect bc_data_query with returnData absent or false, and a schema back rather than rows. This proves the compile path works in isolation, which is the single most useful thing to know before you start debugging a complicated query.

If all three pass, the plumbing is good.

Now go break it

That is the setup done. Six steps, one configuration, no API pages, and an agent that can read your Business Central data by writing AL on the fly.