All guides

AI Assistants & MCP

What is MCP? (Plain-language explainer)

The connector standard that lets AI assistants use ZYLX safely.

2 min readEasy

Why this matters

You'll see the term 'MCP' during assistant setup. You don't need to understand it deeply — but here's the gist, jargon-free.

  1. 1

    MCP in one sentence

    • MCP (Model Context Protocol) is like a universal power adapter between AI assistants and apps: it lets Claude, ChatGPT, or another MCP-capable assistant plug into ZYLX and use a fixed, named set of tools — nothing more.
  2. 2

    Why this is safer than 'giving the AI my passwords'

    • Your assistant never sees your Shopify token or Google credentials. Those stay encrypted inside ZYLX.
    • The assistant only gets the specific tools ZYLX exposes: reading store data, running audits, and submitting suggestions to YOUR approval queue.
    • A read-only connection makes zero durable business changes. Action tools appear or execute only when the connection scope, workspace role, ZYLX policy, and approval state all permit them.
    • Claude, Claude Code, ChatGPT, and Codex use OAuth. Manual bearer tokens remain an advanced developer option and are shown only once.
  3. 3

    What assistants can and cannot do

    • CAN: use the compact 18-tool Business Brain surface, search evidence, read freshness, and compare explicitly authorized businesses.
    • CANNOT: see an ungranted business, turn a member role into owner authority, use an old search result as authorization, or auto-add a future business.