MCP is not the answer to AI in the enterprise.

MCP is not the answer to AI in the enterprise. It is the plumbing. The narrative sounds simple: connect your LLM to ERP via MCP, let it read the data, magic happens.
It will not.

ERP data was built to be processed, not reasoned over. Table names are codes. The same product has different labels in different systems. Business rules are undocumented. Process links are implicit. A purchase order does not know it is step 3 in a chain where steps 1 and 2 already happened. A goods receipt does not know quantities from this source historically mismatch invoices by 20%.

MCP gives access. Access is not understanding.

An LLM on raw ERP will either produce generalities any intern could write, or confident specific answers that are wrong. It treats a table as truth when the truth requires context the table never carried.

Now multiply it. Most enterprises do not run one ERP. They run five. Plus a WMS, a TMS, and three acquisitions that each brought their own stack. Eight MCPs pointing a model at eight systems describing the same reality in different languages, different codes, different hierarchies, zero shared context. That is not eight problems. It is twenty-eight integration seams the model navigates with no map. The complexity is not additive. It is combinatorial.

If the connector was the hard part, SAP would have solved AI ten years ago and we would all be on a beach.

The missing layer is resolution.

Before AI reasons, data needs to carry meaning: standardized names across systems, explicit process context, temporal ordering, codified business rules, lineage to source. The model needs “MATNR” and “item_code” and “product_ref” to be the same thing. It needs to know this vendor’s reliability has declined for six months. It needs to know what should have happened before this step and what should follow.

None of that lives in ERP. None of it arrives through a connector. And it does not emerge from pointing eight pipes at a model hoping it discovers the Rosetta Stone on its own.

MCP solves the last mile. The first mile is the hard part. Almost nobody is talking about it.

That first mile is what we built Maxa to solve. One resolved data foundation where every source is harmonized into standardized business activities with process context, codified rules, and semantic richness LLMs reason over natively. Not another connecting layer.