Model Context Protocol is an open standard that lets an AI model call a vendor's system through a connector. The connector decides which data and actions to expose.

It can read many systems and is strong at analysis and preparation. It cannot hold a resolved identity, store governed definitions, enforce consent across systems or produce an audit record.

A layer that resolves identity across sources and stores each brand definition once, with an owner and a version, so every agent and analyst uses the same meaning.

Skills reduce drift, but they need authoring, maintenance, evals and governance. A skill tells the model what to do, but it cannot and will not stop the model from doing something else.

Because raw results land in context, reconciliation runs on every question, which means nothing persists between users, and long conversations re-read earlier results.

No. Kana uses MCP to bring data in and send actions out, and supplies the shared meaning and control between them that would be missing otherwise.

Kana picks the best model for each task based on quality, cost and latency, so the brand does not have to track it.