{ "mcpServers": { "x1Agent": { "command": "npx", "args": ["-y", "@caterlord/x1-mcp-server"], "env": { "X1_API_ENDPOINT": "https://mcp.x1.tech/mcp" } } } }
Edit your menu from the chat.
Say what you want in plain English. The agent returns a preview — matched items, before/after, risk score — you approve, X1 commits with an idempotency key. No spreadsheet re-keying. No USB-to-shop runs.
Turn a PDF menu into a live menu.
Drop your supplier's PDF, a photo of the wall menu, or a pasted Excel block. The agent extracts a structured draft, validates against your HQ, and classifies every line so you know exactly what will change before you approve.
Ask why — get a real answer.
An item missing from Foodpanda? A shop that won't publish online? One question to your AI. It inspects every surface that can hide an item — category, menu, modifier, shop price, shop availability, online-ordering flag — and comes back with a cause and a safe fix.
Your daily numbers in the same chat.
Ask for trends, compare periods, spot underperformers, catch exceptions — all in the same chat window you already have open. Normalized answers with deltas and caveats, not raw report rows you have to math yourself.
Safe by design.
Every write goes through three modes: Read inspects, Preview diffs and scores risk, Commit requires an approval token and writes with idempotency and audit lineage. No silent writes. Agents never guess IDs — they get disambiguation candidates when names are ambiguous.
Read the API docs. Ship in an afternoon.
REST, webhooks, and the hosted MCP server — one reference with typed SDKs for TypeScript and Python, OpenAPI 3.1, and a sandbox keyed to your account.