BC MCP Server: What Business Central's Model Context Protocol Means for AI Automation

ERP systems have an AI problem. Not that they lack intelligence — they’re packed with business logic. The problem is that AI agents can’t talk to them natively. Every integration is a custom REST API wrapper, a middleware layer, a brittle connector that breaks when someone renames a field. If you’ve ever tried to get an AI assistant to pull live inventory levels or post a sales order into Business Central, you know the pain.

Microsoft’s answer is the BC MCP server — a Model Context Protocol gateway built into Business Central that exposes your ERP data and operations to AI agents through a standardized, self-describing interface. It entered public preview in October 2025 (version 27.1), and it’s quietly becoming one of the most consequential pieces of infrastructure in the Dynamics 365 ecosystem.

This isn’t another Copilot feature bolted onto the UI. This is infrastructure for AI-native automation. Let’s walk through what it actually does, how to use it, and what it means for your business.

---

What MCP Is (in Plain Terms)

Model Context Protocol is an open standard — originally created by Anthropic in late 2024 — that defines how AI applications discover and interact with external data sources. Think of it as USB-C for AI-to-system connections. Instead of every AI tool needing a custom integration for every business application, MCP provides a universal connector.

The architecture is simple:

• The MCP host is the AI application — Copilot Studio, ChatGPT, Claude, GitHub Copilot, or a custom agent you build.
• The MCP server is the gateway exposed by the business application — in this case, Business Central.
• The host discovers what tools are available on the server, understands their inputs and outputs, and invokes them — all through a standard protocol.

The BC MCP server exposes Business Central entities — customers, items, sales orders, vendors, posted invoices — as tools that AI agents can discover, understand, and call. It’s not a chatbot. It’s a communication layer that turns your ERP into something an AI agent can reason about and act upon.

---

What the BC MCP Server Actually Exposes

Out of the box, the BC MCP server provides read-only access to all exposed API pages. That means an AI agent connected to your environment can query customer records, pull inventory levels, retrieve open sales orders, and look up vendor balances — all through standard conversational interfaces.

But read-only is just the starting point. Administrators can configure specific API pages with create, modify, and delete permissions, plus bound actions. Once enabled, agents can:

• Create and update records — Customers, items, sales orders, purchase orders, and more
• Execute business processes — Post documents, change statuses, run business logic
• Take action on behalf of users — With full audit trail showing who performed each operation

Every action respects Business Central’s existing security model. The agent doesn’t get a backdoor. It operates with the authenticated user’s permissions, role-based access controls, and tenant isolation. The audit log shows exactly who did what, when — including when that “who” was an AI agent acting on a user’s behalf.

This is the critical design decision Microsoft made: the MCP server doesn’t bypass security. It extends the same permission model to AI agents that your human users follow. No special privileges. No hidden access.

---

Supported MCP Hosts: It’s Not Just Copilot

One of the most important things to understand about the BC MCP server is that it’s not a Microsoft-only ecosystem play. Microsoft supports three categories of MCP hosts:

1. Copilot Studio — Microsoft’s primary target for the first wave. You create an agent in Copilot Studio, add the Business Central MCP connector, and configure which environment and company to connect to. Takes about five minutes.

2. Visual Studio Code with GitHub Copilot — Developers can connect VS Code directly to the BC MCP server, giving Copilot access to your Business Central data and APIs while coding. This is huge for AL developers who want AI-assisted development with live context from the environment they’re working against.

3. Any MCP-compliant client — Claude, ChatGPT, MCP Inspector, and any custom agent that speaks the MCP specification. Microsoft explicitly supports non-Microsoft hosts. They even document how to register your own Entra ID application for third-party client authentication.

The endpoint is the same for everyone: https://mcp.businesscentral.dynamics.com. You specify the tenant, environment, and company through HTTP headers. That’s it. One gateway. Any MCP host.

---

How It Differs From Standard BC APIs

Business Central has had REST APIs for years. So what does MCP add?

Standard APIs are dumb pipes. The client needs to know the exact endpoint, the field names, the data types, the relationships, and the business rules — all prior knowledge. This is why integrations take weeks to build and break when schemas change.

MCP is self-describing. When an AI host connects to the BC MCP server, it discovers:

• What tools are available (API pages, actions)
• What parameters each tool accepts
• What data types are expected
• What the tool returns
• The schema of every entity

The AI agent doesn’t need pre-programmed knowledge of your Business Central setup. It reads the metadata at connection time and adapts. If you add a custom API page or a new field, the agent picks it up automatically on the next connection. No integration maintenance. No breaking changes when the schema evolves.

This is the difference between shipping a static map to someone and giving them live GPS. Both get them there, but one handles detours.

---

Realistic Use Cases (Not PowerPoint Fantasies)

Let’s ground this in things you’d actually do, not demo-ware.

Customer Service: “What’s My Order Status?”

An AI agent connected to the BC MCP server, surfaced through a Teams chat or a custom portal, can answer questions like:

• “What’s the status of order SO-10432?”
• “When did we last ship to Contoso Ltd?”
• “How much does Alpine Ski House owe us right now?”

The agent reads from the production Business Central environment in real time, not from a cached data warehouse. The customer gets current information. No human opens the ERP.

Sales Team: “Create a Quote for an Existing Customer”

A Copilot Studio agent with write permissions can draft a sales quote from a Teams conversation. The sales rep says “Create a quote for Fabrikam with 50 units of item 1000 at standard pricing.” The agent pulls the customer record, checks inventory, creates the quote, and returns the quote number and total. The rep reviews and confirms. Done in seconds instead of logging into BC and navigating three pages.

Warehouse: “What Needs Picking Today?”

A warehouse supervisor asks a Copilot agent “Show me all unreleased sales orders for today.” The agent queries the MCP server, returns the list, and offers actions: “Release order 10435 for picking?” With write permissions, the supervisor confirms and the order moves to the warehouse shipment flow.

Finance: “Post the Cash Receipt Journal”

The MCP server can execute bound actions on API pages. A finance agent could prepare the cash receipt journal for posting, flag exceptions for review, and — after human approval — trigger the post action. The Journal is posted. The G/L is updated. The audit trail shows the user who approved it and the agent that executed it.

Custom Line-of-Business Agents

This is where things get interesting for custom development. Your company has a unique process — maybe a complex service order workflow, a rental management flow, or a field service dispatch system. With the BC MCP server and a custom development partner, you can build purpose-built AI agents that speak MCP and interact with your Business Central extensions just like they interact with standard entities.

---

The Admin Center MCP Server (2026 Wave 1)

Microsoft didn’t stop at exposing business data. In April 2026 (2026 Release Wave 1), they introduced a second MCP server — this one for the Business Central admin center.

This server exposes environment management APIs as MCP tools. Administrators can ask questions like:

• “What’s the update status for my sandbox environments?”
• “Which environments have extension X installed?”
• “Show me available upgrades for the Production environment.”

And take actions like:

• “Copy Production to a new sandbox”
• “Schedule the next available update window”
• “Investigate that failed upgrade from last week”

This replaces PowerShell scripts and Logic Apps for common admin workflows. The same MCP standard. The same conversational interface. Just pointed at environment management instead of business data.

For partners managing dozens of customer environments, this is the difference between logging into each tenant separately and asking one agent “Give me a health check across all my customers.”

---

Getting Started: What You Need

Setting up the BC MCP server isn’t a consulting engagement. It’s a configuration exercise:

1. Enable MCP in Business Central — Navigate to Feature Management, find the MCP server feature, and enable it for all users. This is a flip-switch operation.

2. Configure which APIs are exposed — By default, all standard APIs are available read-only. To add write operations, use the MCP Configuration page to select specific API pages and assign create/modify/delete permissions and bound actions.

3. Set up your MCP host — In Copilot Studio, add the Business Central MCP connector, authenticate with your Entra ID credentials, and specify environment name and company. For non-Microsoft hosts (Claude, ChatGPT, custom clients), register an application in Entra ID and configure the MCP client with the endpoint URL and authentication tokens.

4. Test — Use MCP Inspector (a free developer tool) to verify connectivity and available tools before building agents.

The prerequisites are minimal: Business Central online version 27.1 or later, proper licensing, and admin permissions to configure the MCP feature and API exposures.

For companies already on Business Central SaaS, this is configuration, not implementation. The server is built in. You’re enabling it, not building it.

---

Limitations and Honest Caveats

Every technology has edges. Here are the ones you should know about the BC MCP server today:

It’s still in public preview. While it’s functional and being used in production by early adopters, Microsoft hasn’t declared general availability yet. That means APIs could change, features could be adjusted, and you should test thoroughly before building mission-critical workflows on top of it.

Write operations need deliberate configuration. Nothing writes by default. If you want agents to create or modify data, an administrator must explicitly enable each operation on each API page. This is the right design, but it means you can’t just “turn on” full agent write access — you need to think through what each agent should be allowed to do.

Performance depends on API design. The MCP server calls your BC API pages. If a custom API page is slow or returns too much data, the agent experience will be slow or truncated. The MCP server doesn’t magically optimize your APIs. Garbage in, garbage out, garbage response times.

AI agents are only as good as their instructions. The BC MCP server provides the tools. The AI host (Copilot Studio, Claude, ChatGPT) provides the reasoning. If your agent’s system prompt is vague, it’ll produce vague results — or worse, confidently wrong ones. Good agent design still matters.

Multi-company and multi-environment complexity. The MCP server connects to one environment and one company per connection. If you need an agent that spans multiple companies (e.g., a consolidated reporting agent across subsidiaries), you’ll need multiple connections or a custom orchestration layer.

Language support. The MCP protocol and server infrastructure are language-agnostic, but the quality of the AI agent’s natural language understanding depends on the host model. If your team operates primarily in Romanian, Spanish, or German, test with your actual language before assuming it works seamlessly.

---

Where This Fits in Your AI Strategy

The BC MCP server isn’t a product you buy. It’s infrastructure you use. The real question is: what do you build on top of it?

If you’re already using Business Central’s built-in AI agents — the Sales Order Agent, Payables Agent, or Expense Agent — the MCP server is the next layer. Those agents are packaged solutions for specific workflows. The MCP server is the open platform that lets you build your own.

The strategy looks something like this:

1. Deploy the built-in agents for high-ROI workflows (sales order processing, AP automation, expense management)
2. Use the BC MCP server to build custom agents for your unique processes — the ones no out-of-the-box agent handles
3. Connect third-party MCP hosts (Claude, ChatGPT, custom tools) when you need AI capabilities beyond what Copilot Studio offers
4. Extend with custom development for API pages that expose your Business Central extensions through MCP

This isn’t a “rip and replace” strategy. It’s additive. You keep your existing Business Central workflows and add AI agents as a new interaction layer — the same data, the same security, the same business logic, accessed conversationally or programmatically by agents that never get tired of looking up order statuses.

---

FAQ

What exactly is the BC MCP server?

It’s a protocol gateway built into Business Central (version 27.1+) that implements the Model Context Protocol — an open standard for AI applications to discover and interact with business data. It exposes Business Central entities like customers, items, and sales orders as standardized “tools” that AI agents (Copilot Studio, ChatGPT, Claude, custom agents) can discover and use without custom integration code.

Do I need to be a developer to use it?

To enable and connect Copilot Studio, no — it’s configuration, not development. A system administrator flips the feature on, configures which APIs are exposed, and connects the host. For custom AI agents, non-Microsoft hosts, or exposing custom BC extensions, you’ll need development work — but the protocol itself handles the heavy integration lifting.

Is the BC MCP server secure?

Yes, and by design. The MCP server does not bypass Business Central security. All operations execute with the authenticated user’s identity and are subject to the same permissions, role-based access controls, and tenant isolation as any other BC operation. Additionally, write operations are disabled by default and require explicit administrator configuration. Every action — including AI agent actions — is logged in the BC audit trail.

Can I use it with ChatGPT or Claude instead of Copilot Studio?

Yes. Microsoft’s documentation explicitly supports non-Microsoft MCP hosts including ChatGPT and Claude. You’ll need to register an application in Microsoft Entra ID for authentication (Microsoft hosts use a pre-registered app), but the connection process is documented and straightforward.

What’s the difference between the data MCP server and the admin center MCP server?

There are two separate MCP servers. The data MCP server (introduced October 2025) exposes Business Central business data — customers, items, orders, journals — to AI agents. The admin center MCP server (introduced April 2026) exposes environment management APIs — update status, installed extensions, environment copies — to AI agents. They use the same MCP standard but serve entirely different purposes: one for business operations, one for system administration.

What does it cost?

The BC MCP server is included with your Business Central subscription. There’s no separate licensing for the server itself. However, using AI hosts like Copilot Studio may require Copilot capacity or Microsoft 365 Copilot licenses depending on your configuration. For non-Microsoft hosts (ChatGPT, Claude), you pay whatever those platforms charge for API usage. The MCP server doesn’t add a per-call cost.

Can I expose my custom BC extensions through MCP?

Yes. Any API page in Business Central — standard or custom — can be exposed through the MCP server. If you’ve built custom entities, custom workflows, or AL extensions, you can make them available to AI agents by adding the relevant API pages to your MCP configuration. This is where custom development creates the bridge between your unique business processes and AI automation.

---

What’s Next

The BC MCP server is infrastructure, not magic. It won’t automate your business by itself. But it removes the single biggest barrier to AI-powered ERP automation: the integration tax. No more building REST wrappers for every AI tool. No more maintaining brittle connectors. One protocol. One endpoint. Any AI agent.

If you’re already on Business Central, enable the MCP server and connect it to Copilot Studio. Start with read-only queries — “show me open orders for customer X” — and build from there. The learning curve is shallow, and the value compounds as you layer on more capabilities.

We help companies set up, configure, and extend the BC MCP server — from enabling the feature and exposing the right APIs to building custom agents and custom API pages for your unique processes.

Book a 30-minute call →