Using Claude (or any AI assistant) with Business Central through MCP
With the MCP server, an assistant such as Claude, Copilot or others can work against live Business Central data with your permissions. It is very good at some things and should not be trusted with others. Practical uses, where to draw the line, and how to set it up so it helps without becoming a risk.
The Business Central MCP server means an AI assistant you already use can connect to Business Central and work with live data, within the permissions you give it. Claude, Microsoft's own Copilot and other MCP-compatible assistants can all do this. It's the closest thing yet to having an analyst who can query the system for you in plain English. Used well, it's a big time-saver. Used carelessly, it's a way to make confident mistakes faster.
What it's good for
Questions about your data. "Which customers have bought less this quarter than last?" "What's our stock exposure on this product range?" "Which suppliers have we paid late most often this year?" It reads the data and summarises it.
Investigations. "Why doesn't the debtors control account match the aged receivables?" It can pull the entries and help you find the ones that don't belong (see direct posting).
Month-end checklists. Working through a list: bank reconciled, accruals posted, no unposted documents, intercompany balances agree, and telling you what's left.
Drafting. Commentary for the management accounts, an email to a customer about their account, a summary of a supplier's history.
Explaining. "What does this posting group setup mean?" with reference to your actual configuration.
Where to draw the line
Don't let it post to the ledger, change setup or change bank details. Give it read access for analysis; keep writing with people or approved agents with narrow permissions.
Check its arithmetic on anything that matters. Ask it to show the entries behind a figure, and spot-check them.
Don't treat it as the source of truth. Business Central is. The assistant is a way of reading it.
Connect with an account whose permissions are read-only for the relevant areas.
Try it on a sandbox first.
Be thoughtful about personal data: the assistant's provider is processing it (AI and data protection).
The result is a finance team that can ask the system questions in their own words and get answers in seconds, with the system's own security still in charge.
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.