Weekend Project: Teaching AI to Talk to Business Central

What happens when you spend a weekend connecting Claude to Dynamics 365 Business Central?

The Setup: From Zero to Connected

This weekend, I went down the rabbit hole of connecting AI to Business Central using the Model Context Protocol (MCP). MCP is essentially a “USB port for AI” — a universal standard that lets AI assistants talk directly to business systems.

The setup involved three key pieces:

1. Azure AD App Registration — Creating an app with Application permissions (not Delegated — learned that the hard way!) You need API.ReadWrite.All and Automation.ReadWrite.All permissions, plus admin consent.

2. Business Central Configuration — This is where BC’s new MCP capabilities shine:

• Feature Management — Search for “Feature Management” and enable “Enable MCP Server access” for all users

• Azure Active Directory Applications — Register your Azure AD app’s Client ID and assign permission sets (like D365 BUS FULL ACCESS for testing)

• MCP Server Configurations — The new page where the magic happens:

  • Create a configuration (e.g., “Sales Agent”)

  • Click “Add all standard APIs as tools” — this automatically exposes customers, items, orders, vendors, and more

  • Set Active = Yes to enable it

  • Optionally enable Dynamic Tool Mode” for auto-discovery of 50+ APIs

  • Toggle “Unblock Edit Tools” to allow create/update/delete operations (careful in production!)

3. Python MCP Server — A lightweight server that translates AI tool calls into BC API requests. It handles OAuth token caching, company ID lookups, and OData query building. More on this in a future blog post.

The result? Instead of navigating menus and running reports, I can now ask questions in plain English and get answers from live BC data.

AI Agents: Specialized Assistants for Every Role

Beyond basic queries, I’m planning to generate agents for different business functions:

Agent	Purpose
Sales Manager Agent	Sales performance, quote tracking, customer analytics
Payables Agent	A/P management, vendor balances, payment planning
Fixed Assets Agent	Asset tracking, depreciation schedules
Jobs Agent	Project costing, budget vs actual analysis

Each agent will understand its domain deeply and answer role-specific questions like “Who are my top 10 salespeople by quotes this month?” or “What invoices are due this week?”

Included below are a few examples of interrogating the Payables Agent:

Example 1: Reading Data — “What Vendors Do I Owe Money To?”

MCP Tool Call: bc_list_aged_ap

{

  "top": 10

}

```

Result:

Vendor	Balance Due	Aging
Wide World Importers	$18,193.10	90+ days
Nod Publishers	$17,273.87	90+ days
Graphic Design Institute	$6,979.57	90+ days
First Up Consultants	$4,903.88	90+ days
Fabrikam, Inc.	$2,082.13	90+ days

One question. Instant answer. No report navigation required.

Example 2: Writing Data — Creating a Sales Quote

But reading data is just the beginning. The real power is when AI can create transactions. Here’s a complete example of creating a sales quote through natural conversation.

Prompt: Create a sales quote for Adatum with 2 Athens desks

Step 1: Find the Customer

MCP Tool: bc_list_customers

`json

{

  "filter": "contains(displayName,'Adatum')",

  "top": 5

}

API Response:

```json

{

  "value": [{

    "id": "92f5e1bd-1b4e-ef11-bfe7-6045bde9c989",

    "number": "10000",

    "displayName": "Adatum Corporation",

    "city": "Atlanta",

    "state": "GA"

  }]

}

```

Step 2: Find the Item

MCP Tool: bc_list_items

```json

{

  "value": [{

    "id": "92f5e1bd-1b4e-ef11-bfe7-6045bde9c989",

    "number": "10000",

    "displayName": "Adatum Corporation",

    "city": "Atlanta",

    "state": "GA"

  }]

}

```
```json

{

  "filter": "contains(displayName,'Desk')",

  "top": 10

}

```

API Response:

```json

{

  "value": [{

    "id": "7294b9c2-1b4e-ef11-bfe7-6045bde9c989",

    "number": "1896-S",

    "displayName": "ATHENS Desk",

    "unitPrice": 1000.8,

    "inventory": 4

  }]

}

```

Step 3: Create the Sales Quote

MCP Tool: bc_create_sales_quote

```json

{

  "customerNumber": "10000"

}

```

API Response:

```json

{

  "id": "1a59f93f-3500-f111-8405-6045bda74b53",

  "number": "S-QUO1005",

  "documentDate": "2026-02-02",

  "customerNumber": "10000",

  "customerName": "Adatum Corporation",

  "status": "Draft"

}

```

Step 4: Add Line Item

MCP Tool:** bc_add_sales_quote_line

```json

{

  "quoteId": "1a59f93f-3500-f111-8405-6045bda74b53",

  "itemNumber": "1896-S",

  "quantity": 2

}

```

API Response:

```json

{

  "id": "0397fd45-3500-f111-8405-6045bda74b53",

  "lineObjectNumber": "1896-S",

  "description": "ATHENS Desk",

  "quantity": 2,

  "unitPrice": 1000.8,

  "amountExcludingTax": 2001.6,

  "taxPercent": 6.0002,

  "totalTaxAmount": 120.1,

  "amountIncludingTax": 2121.7

}

```

The Road Ahead

While this proof-of-concept is exciting, there’s still work to be done. More investigation is needed to understand the full capabilities and limitations of BC’s native MCP endpoint. More tests are needed to integrate custom APIs … More exploration is needed to learn how to make this accessible to regular business users — not just developers comfortable with VS Code, Python and JSON.

The real opportunity lies in integrating these AI capabilities with the tools people already use.

This is just the first in a series of blog posts documenting this journey. In upcoming posts, I’ll dive deeper into specific agents, explore the quote-to-cash flow, and share lessons learned along the way.

And maybe the future is not learning the ERP user interface but AI tools to talk to ERP directly via exposed APIS without clicking through menus.

Stay tuned.