ERP systems MCP servers

What is an MCP server for ERP systems?

An MCP server for an ERP system connects AI agents to your ERP instance via the open Model Context Protocol, whether that is SAP S/4HANA, Oracle, Microsoft Dynamics 365, abas or proALPHA. The AI agent gets standardised, controlled access to orders, stock, master data, suppliers, customers and postings, within clearly defined tools and permissions.

In mid-sized companies, ERP systems are often the most valuable but also the most complex data asset. Anyone who wants to use AI agents seriously in the company cannot avoid the ERP connection, and this is where MCP is at its strongest: instead of separate connectors for each AI model, one MCP server is enough, and Claude, GPT-5, Gemini or Microsoft Copilot can all connect to it equally.

This overview page gives a practical summary. For SAP we already have a detailed individual guide: MCP server for SAP. There we also put the SAP API Policy into context, which prohibits AI agents from using SAP interfaces outside SAP-approved routes and therefore has to be clarified before any SAP project. Here we look at the broader landscape, with a focus on the German mid-sized companies (abas, proALPHA), where SAP is not always the obvious choice.

The ERP MCP landscape in 2026 compared

Not every ERP offers the same interfaces. The table below gives a rough classification. For a concrete feasibility check we need your system version, customising status and hosting model anyway.

ERP-specific use cases

1. Order and delivery status by voice "What is the status of customer X's orders?" The agent pulls from the ERP: open orders, planned delivery dates, blocked line items, dunning status. Ideal for sales, internal sales support and customer service, whether SAP, Dynamics or abas.
2. Stock and warehouse analytics Slow-mover analysis, availability checks, safety stock: the agent pulls the data from the ERP and delivers concrete suggestions: "These 12 items have had no movement for 90+ days, here is their stock value".
3. Purchase proposals and supplier routing The agent compares purchasing needs against supplier terms, lead times and minimum quantities, and creates purchase proposals in the ERP. Purchasing reviews and approves (always a four-eyes check for ERP write operations).
4. Master data maintenance New customer from the CRM, new item from the PIM, new supplier: the agent creates the master data consistently in the ERP, checks for duplicates and assigns number ranges according to the rules. This takes the load off manual master data maintenance.
5. Management accounts, revenue and cost reports From ERP data (FI/CO in SAP, GL in Dynamics, management accounts in abas/proALPHA), the agent builds reports at the push of a button or by voice query: "How was revenue with the top 10 customers in Q1 vs. the prior year?"
6. Order confirmations, reminders and transaction emails Routine communication around ERP transactions: order confirmation email with correct delivery dates, dunning levels according to the dunning rule, status updates to customers. Proposed by the agent, approved by the employee.

German mid-sized companies: abas, proALPHA and the "non-SAP" world

Among mid-sized companies in the DACH region, abas ERP (based in Karlsruhe, focused on machine and plant engineering and industrial manufacturing) and proALPHA (Weilerbach, very strong with manufacturing mid-sized companies) are the obvious choice over SAP for many businesses. Both vendors have modernised their interface strategy in recent years:

• abas: REST API with OAuth2, EDP interface (classic), abas Connect for modern integrations. The data model is idiosyncratic (tables with hierarchies) but well documented.
• proALPHA: OData interface, REST workbench and the Integration Server. Here too, a solid foundation for MCP connections, especially if proALPHA already runs in the cloud variant (proALPHA Cloud).
• Microsoft Dynamics 365 Business Central: Widespread among mid-sized companies in the DACH region, often together with Dataverse and Power Platform. Microsoft ships its own MCP server (endpoint mcp.businesscentral.dynamics.com, generally available since 1 April 2026 according to the release plan). Details in the guide MCP server for Business Central.
• Microsoft Dynamics 365 Finance & Operations: Here too there is an official MCP server from Microsoft, the dynamic "Dynamics 365 ERP MCP" server. It works with the security roles of the signed-in user. Details in the guide MCP server for Dynamics 365.

For family-owned businesses and manufacturing mid-sized companies in particular, which cannot simply "swap out" their ERP, MCP is the pragmatic way to unlock AI value from the existing system, without an ERP migration.

Security patterns for ERP MCP

ERP holds the most critical data in your company. Our security standard:

• EU or on-premises hosting for a dedicated MCP server: with SAP often behind the Cloud Connector, with on-premises ERP inside the company network behind a VPN. With Microsoft's official MCP server, Microsoft runs the server itself.
• Dedicated service user in the ERP with minimal rights per use case (e.g. "read-only on master data + order status").
• Four-eyes principle for write operations: orders, postings and master data changes are, as a rule, proposed by the agent and approved by employees.
• Audit logging: Every tool call documented: timestamp, tool, affected ERP document, employee who triggered it. This is the basis for audit-ready evidence.
• DPA with the AI provider and EU data residency (Claude via AWS Bedrock in an EU region, Azure OpenAI in an EU region, etc.) or a local model.
• Works council: We provide a template for information / co-determination, relevant wherever employee data is affected.

What does it cost?

• Use-case-oriented approach: First one clearly scoped use case (e.g. master data lookup or order status bot) in production, then expansion.
• First results with a clearly scoped use case are realistic. ERP projects take longer than, for example, CRM or accounting connections.

Frequently asked questions

Which ERP systems can be connected via MCP?

SAP S/4HANA and ECC, Oracle E-Business Suite/Fusion, Microsoft Dynamics 365 Business Central and Finance & Operations, abas, proALPHA, Sage 100/X3, Infor, and many other ERP systems with a documented API or OData interface. Whether direct database access is permitted depends on the vendor's terms; for SAP, the API Policy relies on published interfaces and additionally restricts AI agents.

How does SAP MCP differ from Dynamics MCP?

The biggest difference lies in the vendor rules. For AI agents on SAP, the SAP API Policy applies, which prohibits such agents outside SAP-approved routes; technically, connections run via BAPI/RFC, OData and the SAP Cloud Connector. Microsoft offers its own MCP servers for Business Central and Dynamics 365 Finance & Operations; REST/OData APIs and the Power Platform are also available. You can find the in-depth guide at MCP server for SAP.

Are abas and proALPHA suitable for MCP?

Yes, both. abas offers a REST API and EDP interfaces, proALPHA has OData and an Integration Server. Both can realistically be connected via an MCP adapter, which for manufacturing mid-sized companies in particular is a good lever to unlock AI value from the existing system.

Can the AI agent create orders or postings?

Technically yes. In practice, for ERP write operations we recommend a four-eyes principle: the agent prepares, a human approves. That still saves most of the time while providing the necessary control.

What about GDPR and data residency?

A dedicated MCP server runs in the EU, and for on-premises ERP usually inside the company network or behind a VPN/Cloud Connector. If you use Microsoft's official MCP server for Business Central or Dynamics 365, Microsoft runs the server itself; the storage location then follows Microsoft's specifications for your environment. As AI providers, models with EU data residency come into question (Claude via AWS Bedrock in an EU region, Azure OpenAI in an EU region) or a local model. A DPA is standard.

How long does an ERP MCP project take?

For a first productive use case (reporting, master data lookup, order status bot) we plan more time than for CRM or accounting connections. More complex write workflows take correspondingly longer. We recommend thinking in use-case-sized steps rather than "big bang".

Is on-premises ERP a problem?

No. The MCP server itself can run on-premises or in a private cloud and access the ERP through secure channels (VPN, SAP Cloud Connector). For many mid-sized companies, that is exactly the clean solution.

Which AI models are supported?

All MCP-compatible ones: Anthropic Claude, OpenAI GPT, Google Gemini, Microsoft Copilot. MCP decouples model and tool, so you can switch the model at any time without rebuilding the ERP integration.