Microsoft AI
The layer between a language model and your ERP.
Four kinds of Microsoft AI work, one common thread: a language model is only useful in a business when it can reach real data without being able to break anything. MCP is the foundation, and it ships today. Copilot, Copilot Studio and Azure AI build on the same access layer.
The stack
Four services, honest statuses.
MCP engineering
Shipping today
MCP for Business Central
The Model Context Protocol layer that lets Claude, Copilot or VS Code read and act on your BC data — scoped, audited, running against the Microsoft-hosted BC MCP server.
Microsoft 365
In scoping
Microsoft Copilot
Custom Copilot connectors and an adoption plan, so the licenses you already pay for reach past Outlook and Teams — into BC, SharePoint and vendor data.
/microsoft/copilot
Agents
In scoping
Copilot Studio
A custom agent scoped to one well-defined job, with an MCP-backed action layer where it needs to write back into BC or your ticketing.
/microsoft/copilot-studio
Azure
In scoping
Azure AI
Tenant-grounded assistants on Azure AI Foundry — retrieval over your SharePoint, BC and document estate, with EU data residency and your existing Entra identity.
/microsoft/azure-ai
The plumbing underneath is open source: bc-mcp-proxy, MIT-licensed on GitHub.
How they fit
MCP first. Everything else builds on it.
A Copilot connector, a Studio agent and an Azure AI assistant all hit the same wall: they need governed access to Business Central. That access layer is what we build first, and it's the same layer whichever front-end you put on top. Start there, and the rest becomes an upgrade rather than a rebuild.
- Microsoft Partner
- 20+ years on Business Central
- Shopify Expert
Not sure which layer is the right next step?
Book a short call. We'll look at your stack and tell you honestly which of these is worth your money now — or whether to wait a year.