MCP (Model Context Protocol): How AI Uses Your Business Data
Your AI only knows what you paste into the chat. Once the conversation ends, that context is gone. MCP (Model Context Protocol) takes a different approach: it connects AI models directly to your data in real time, including your Xentral system.
Key Takeaways
- MCP is an open standard: it connects AI models directly to external systems and data sources
- Xentral MCP Server: we make your ERP data available live to Claude and ChatGPT
- More context for AI: your AI answers questions about orders, inventory, customer data, and KPIs straight from your system
- Agent Hub: lets you actively drive AI agents, not just pull data
What is MCP (Model Context Protocol)?
MCP stands for Model Context Protocol, an open standard that gives AI models structured access to external applications and data sources. Instead of manually copying information into a chat, the AI pulls the relevant data straight from the system where it lives.
Anthropic developed the standard in November 2024 and released MCP as an open specification. The code and documentation are freely available on GitHub and at modelcontextprotocol.io. Shortly after launch, other AI providers like OpenAI and Google DeepMind adopted the standard as well.
With that, AI providers solved a problem that kept getting bigger as more AI applications and systems came online: without a shared standard, every combination of AI app and system needed its own custom-built interface.
That's manageable with a handful of apps and systems. At scale, the effort quickly gets out of hand. MCP replaces this patchwork of one-off interfaces with a single standardized connection. Each system builds one MCP Server, each AI application builds one MCP Client, and the two work together with no extra tweaking.
Before MCP, working with AI tools usually looked like this: take a screenshot, export a table, paste in some text. Every answer was only as fresh as the last thing you pasted. MCP fixes that by creating a direct, standardized connection between the AI and your system.
One thing to keep in mind: MCP isn't a Xentral product. It's a vendor-neutral standard. Software providers across many industries, not just ERP, are building their own MCP Servers to open up their systems to AI applications.
What is an MCP Server?
An MCP Server is the side of the connection that exposes a system's data and functions to AI applications. In the case of an ERP, that could also apply to project management tools or code repositories. It makes the system readable for the AI apps it's connected to and, depending on the permissions you set, controllable as well.
The counterpart to the MCP Server is the MCP Client: the AI application that connects to the server and uses its data and functions, such as Claude or ChatGPT. An MCP Server can be connected to several clients at the same time, and a client can talk to several servers in return.
How does an MCP Server work?
With MCP, you're dealing with two roles:
- The MCP Server exposes a system's data and functions, for example your ERP's.
- The MCP Client is the AI application that pulls and uses that data, such as Claude or ChatGPT.
Technically, MCP runs on a host-client-server architecture. The AI application you interact with is the host. The host spins up its own client for each connection, and each client talks to one server. Communication between client and server runs over JSON-RPC 2.0, a well-established, text-based protocol for requests and responses.
So the AI knows what it can pull from a server, MCP defines three fixed building blocks:
- Tools: executable functions the AI can call on purpose, such as an inventory lookup
- Resources: structured data sources the AI can read, such as a customer list
- Prompts: ready-made templates for recurring tasks that the server provides to the AI
When you ask your AI a question, the client sends the request to the right MCP Server. The server returns the current data live, no export step in between. Your AI answers based on what's actually in your system at that moment. It gets more context about your business as a result.
On security, MCP relies on explicit permissions: every access to a tool or a resource requires your approval, nothing runs automatically in the background. Servers run locally by default, and external access has to be granted separately.
The Xentral MCP Server: MCP in the Context of Your ERP
The Xentral MCP Server opens up your Xentral system to supported AI applications. Once connected, the AI can access the following, depending on the permissions you've granted:
- Customer and product data
- Sales orders and delivery notes
- Inventory levels and stock counts
- KPIs, dashboards, and business plan content
- Invoices and open items
The Xentral MCP Server's functions are grouped into three areas:
- The first area covers ERP configuration, such as business model and system settings.
- The second area covers day-to-day operations in read-only mode. At the center of it: the Xentral Co-Pilot, which lets you ask questions about customers, orders, products, invoices, delivery notes, and returns in plain language.
- The third area covers day-to-day operations in write mode, meaning creating and updating records. This third area is only available to customers with access to the Agent Hub.
What Can You Do with the Xentral MCP Server?
The Xentral MCP Server's functions are grouped into three areas:
- ERP configuration: view your business model and system settings
- Day-to-day, read-only: use the Xentral Co-Pilot to ask questions in plain language about customers, orders, products, invoices, delivery notes, and returns
- Day-to-day, write access: create and update records, for example tasks, CRM entries, or emails. Available only with access to the Agent Hub.
You can check inventory, order status, or open invoices, pull reports on revenue, margin, or top-selling products, or get help making sense of operational data.
With access to the Agent Hub, you take it a step further and let the AI create tasks, update CRM entries, or trigger follow-ups on its own.
Use Cases for the Xentral MCP Server
Ask how much revenue came in through Shopify last week and what an ad campaign cost you. Your AI answers both in one go.
Pull current stock for a product or product group without switching over to the system.
Have your AI put together analyses on revenue, margin, top-selling products, or open items straight from your live data.
Ask about the status of individual orders or delivery notes without manually searching.
Through the API reference built into the MCP Server, you get answers about integrations and automations based on the real Xentral endpoints.
Let the AI create tasks, update CRM entries, or trigger follow-ups, right from your AI interface.
10 Sample Prompts for Your Xentral MCP Server
Here are 10 sample prompts split into different categories to get you started.
Customer Insights
Prompt 1 (Customer value analysis):
"Pull my Xentral customer data. Who are the top 10 customers by total order volume over the last year, and when did they last place an order?"
Prompt 2 (Identify inactive customers):
"Which customers haven't placed an order in the last 6 months even though they previously generated more than €2,000 in revenue? Show me name and email so I can build a reactivation campaign."
Orders
Prompt 3 (Order lead times):
"Analyze orders from the last 30 days. What's the average time between order receipt, delivery note creation, and invoice date?"
Prompt 4 (Compare sales channels):
"Group all orders from last quarter by their source channel. For each channel, show me the number of orders, total revenue, and average order value."
Products & Inventory
Prompt 5 (Stock coverage & bestsellers):
"Compare current inventory levels with sales figures from the last 30 days. Which top 5 products are at risk of selling out in the next two weeks?"
Prompt 6 (Slow-mover analysis):
"Identify products where we currently have more than 50 units in stock but have sold fewer than 5 in the last 90 days. Show me SKU, name, and current inventory value."
Invoices & Financials
Prompt 7 (Open items):
"Give me an overview of all open invoices where the due date is more than 14 days overdue. Sort by highest outstanding amount and include customer names."
Prompt 8 (Margin analysis):
"Based on invoices and KPIs from the last month, calculate the average gross margin (revenue minus cost of goods) per product category."
Cross-Platform Live BI (Zero Exports)
Prompt 9 (Real-time ROAS from Xentral & ad managers):
"Using the relevant MCP connections, pull ad spend from Google Ads and Meta Ads for the last week. Compare that live spend directly against actual revenue and margin from our Xentral invoices for the same period. What's our real combined ROAS?"
Prompt 10 (Amazon performance & ERP sync):
"Query our sales figures from the last 7 days via the Amazon MCP Server and compare them with the Amazon orders imported into Xentral. Are there any discrepancies in unit counts or incomplete order transfers?"
Open Beta and Agent Hub Closed Beta: the Two Access Levels
Xentral currently runs two parallel betas for AI access to your system. Which beta you're on determines whether you get read-only access or read plus write access.
MCP Server Open Beta | Agent Hub Closed Beta | |
Availability | All Xentral users, since mid-July 2026 | Rolling out gradually to 50 to 100 customers at a time |
Access level | Read-only: ask questions, analyze data | Read and write: drive agents directly |
Scope | Customer, order, and inventory data, KPIs, dashboards, business plan content | Service agents, process agents, custom agents, and the Co-Pilot |
Timeline | Ongoing, no gated onboarding | Runs until August 31, 2026, then general availability |
With MCP Server access, you analyze and query your data. Only with additional access to the Agent Hub do you actively drive agents: create tasks, update records, trigger follow-ups, straight from the AI interface.
Benefits of the Xentral MCP Server
- Live data instead of exports: Your AI reads orders, inventory, customer data, and invoices straight from Xentral. No exports, no stale numbers.
- Broader context for better answers: Alongside transactional data, your AI also sees KPIs, dashboards, and business plan content, meaning the same operational context a team member would use.
- Cross-platform analysis: Connect your AI to other tools like ad platforms, and you can reconcile ad spend directly against live revenue and margin from Xentral.
MCP vs. Classic AI Usage
Plenty of teams already use Claude or ChatGPT day-to-day. The difference with classic chat usage comes down to access: an AI without an MCP connection only knows what you paste into the chat or attach as a file, and only for that one conversation.
With the Xentral MCP Server, you connect your AI to your system on an ongoing basis. The connection stays live, the data is current with every request, and you don't have to manually feed anything in.
Who Can Use the Xentral MCP Server?
The Xentral MCP Server works with several AI applications. Currently supported: Claude Code, Claude Desktop, and ChatGPT (Plus, Pro, or Business). Support for other MCP-capable applications like Cursor and Windsurf is on the roadmap.
To set it up, you need:
- An active Xentral account
- The right permissions: with administrator rights, all the necessary access is granted automatically. Otherwise, standard views like Addresses, Orders, Delivery Notes, and Products, plus the Analytics Platform, have to be assigned to you manually.
- A subscription with Anthropic (for Claude) or OpenAI (for ChatGPT), depending on which application you connect
The first time you connect, you sign in once using your regular Xentral credentials. Sign-in runs through OAuth, so you don't have to manage your own API access tokens.
Set Up the Xentral MCP Server and Get Started
Connect your AI assistant to your Xentral system in just a few steps. Our documentation walks you through everything you need for setup, authentication, and your first prompts.
Conclusion: MCP Server Gives Your AI the Context It Needs
The Xentral MCP Server connects your AI directly and live to your ERP, without exports, without copy-paste. Today, you'll mostly use it for queries and analysis. With additional access to the Agent Hub, you take it a step further and drive agents straight from your AI interface.