A common frustration with older ERP systems is that you can't point a modern AI tool at your own data. The system is closed, so questions your team would love to ask, about customers, sales, or open orders, stay locked behind whatever reports the vendor built years ago. When you're choosing a new platform, the ability to connect AI to your data is worth checking before you commit, because it's a hard limit you'll live with for years.
Business Central answers this two ways, and they're different tools for different jobs. One is the Copilot sidecar built into the application. The other is a native MCP server that lets an outside assistant like Claude or Microsoft Copilot reach into your data. Knowing which does what keeps expectations straight.
The Copilot Sidecar
Inside Business Central there's a Copilot panel you open from an icon at the top right. Ask it questions about your data, where to find something, how to run a process, or how to complete a task in the system, and it answers from within the application. It reads as a knowledgeable assistant sitting alongside your work. What it won't do is complete operations for you. It helps you find and understand; it doesn't post the journal.
The MCP Server
MCP, the Model Context Protocol, ships natively with Business Central. Setting it up is a configuration step, not a separate purchase, and there's no added fee for turning it on. Once it's enabled, you can connect either Claude or Microsoft Copilot to your Business Central environment and work against the tables you've exposed.
The difference from the sidecar is that this connection can actually do work. Through Claude or Copilot over MCP, you can create, read, update, and delete records against the enabled tables. That means keying a journal entry from a conversation, or having the assistant reconcile a bank account by reviewing and matching transactions and only asking you when it can't classify something on its own.
A customer sends a complicated order in a spreadsheet without your product IDs. An assistant connected over MCP can analyze the sheet, match items to your product library, and key the sales order, provided your product records carry the attributes needed to identify what each line is asking for.
What Has to Be in Place
The write-back capability is only as good as the data behind it. Matching an inbound order to your catalog depends on a product library set up so items can be identified by their attributes. If your product records are thin, the assistant has nothing to match against. The same discipline that makes any AI useful on your data, clean and well-structured records, applies here.
- Copilot sidecar answers questions and guides processes, but does not complete operations
- MCP server, configured once with no extra fee, connects Claude or Copilot to your data
- Over MCP, an assistant can create, read, update, and delete records on enabled tables
- Write-back against your catalog depends on product records structured by attribute
If an ERP evaluation is stalling on whether the platform will let you connect the AI tools your team already wants to use, this is the capability to test in a demo rather than take on faith: ask to see a real query against real data, and ask to see the assistant key a transaction over MCP.