Email this post as a PDFAI and agents · 4 November 2026 · 2 min read

Configuring the Business Central MCP server safely

The Business Central MCP server lets AI agents built outside BC read and act on your data. That is powerful, and it is a door into your ERP. Which API pages to expose, which permissions to give, how to register clients in Entra, what to log, and the checklist to run before you switch anything on.

Contents
  1. 1. Expose only what's needed
  2. 2. Narrow permissions
  3. 3. Register clients properly
  4. 4. Environments
  5. 5. Log and look
  6. 6. Data protection
  7. The checklist before switching on

The Business Central MCP server (generally available since April 2026, at mcp.businesscentral.dynamics.com) lets agents in Copilot Studio, Visual Studio Code and registered third-party clients discover and use Business Central data and actions, under Business Central's own permissions. It's the most important AI capability in the platform, and it deserves the same care as opening any integration into your ERP. (See agentic AI in Business Central for the background.)

1. Expose only what's needed

You choose which API pages the MCP server exposes in a configuration, and static validation checks it before it's used. Start with the minimum for the use case: a credit control agent needs customers and customer ledger entries, not vendors, G/L or setup tables. Prefer read-only pages. Expose write operations only where a specific, approved agent needs them. Keep separate configurations for separate purposes rather than one that exposes everything.

2. Narrow permissions

The MCP server acts with the permissions of the identity the agent uses. That permission set is the real control. Create one per agent or use case, with read access to exactly the tables needed, and no posting, no setup, no bank details, no SUPER. If an agent must create something, give it insert on that table only, and keep posting with a person. (See permissions.)

3. Register clients properly

Copilot Studio and Visual Studio Code are pre-authorised. Third-party MCP clients must be registered in Microsoft Entra ID. Register each one separately with a clear name, so you can see and revoke it individually. Don't share one registration across tools. Use OAuth as designed; don't build workarounds with stored passwords.

4. Environments

Try everything in a sandbox first (environments). Point development agents at sandboxes only. Production access is a separate, approved step.

5. Log and look

MCP calls are logged with success or failure, and configuration changes are audited through Microsoft Purview. Make sure someone looks at the logs: unusual volumes, failures, access outside working hours, new clients. Telemetry in Application Insights gives you more (telemetry).

6. Data protection

Anything the agent reads may end up in its context or memory, outside Business Central. Think about personal data before you expose it, and record the processing (see AI and data protection).

The checklist before switching on

  1. Named business owner for the use case.
  2. Configuration exposes only the API pages needed; read-only by default.
  3. Dedicated permission set, reviewed; no posting, no setup, no bank data.
  4. Client registered individually in Entra.
  5. Tested in a sandbox.
  6. Logging confirmed; someone named looks at it.
  7. Data protection considered and recorded.
  8. A way to switch it off quickly, and someone who knows how.
Email this post as a PDF
A branded PDF of this post, sent to your inbox to keep, print or share. Copy in colleagues if you like.
We only use these addresses to send this PDF. Nobody is added to a mailing list. Mike at Business Boxed receives a copy. See our privacy notice.
Email this post
Talk to us
Want this done in your Business Central?
Book a 30 minute callGet the monthly newsletter

Related posts