Skip to main content
Ready for Commerce has an MCP server, the open standard AI assistants use to call other apps’ tools. Its address:
Connected to it, an AI assistant such as Claude, ChatGPT, Gemini, Cursor or VS Code reads and works with your organization’s data through the same operations as this API, as you, within what you allow.

Connect your AI

1

Open Connect an AI

In the app, go to Settings › API and connections and press Connect an AI. Choose your AI: the window adds the address by a button where your AI has one, gives Claude Code a command to paste, and otherwise shows the address and where to paste it. Any AI that takes an MCP server’s address can connect, from its own settings.
2

Sign in and choose what it may do

Your AI opens Ready for Commerce’s sign-in, then one page that asks which organization it works with, when you belong to more than one, and what it may do there: Read only, chosen first, Full access or Custom.
3

Allow access

Your AI now lists Ready for Commerce’s tools. Ask it, for example, how many exports ran this week.
Your AI acts as you. Each of its calls is checked as yours, read live: it never does more than your own role in that organization allows, even with Full access, and its connection ends when you leave the organization or reset your password.

Disconnect an AI

Settings › API and connections › Connected AIs lists the organization’s connections. Each person disconnects their own, and a person allowed to manage connections disconnects anyone’s. A disconnected AI stops at once, until someone connects it again.

The tools are the API

Each tool is an operation of this API, with the same checks, permissions and errors: get_org is Get the organization, and start_export is Start an export. A few reads of one record share a tool, which a method argument chooses: get_job reads a job, its messages or a file link. What only a person does in the app is never a tool.
  • A write takes an optional idempotency_key, sent as the Idempotency-Key of its call, so an AI that retries never does a write twice.
  • An error comes back as the tool’s result, with its code and a sentence that tells the AI what to do, such as how long to wait or which permission it lacks.

Limits

  • Calls: an AI connection makes calls a minute, reads and writes together, and bulk calls, inside its organization’s budget (Rate limits).
  • Pages: a tool that lists answers rows unless the AI asks for more, up to , and nextCursor goes on from there. An answer too long for an AI to read is cut to the rows that fit, with truncated: true.

No personal data reaches an AI

A field that holds personal data never reaches a tool, and an export an AI starts leaves out the columns that name people, such as who created or last changed a record.

Call it from a program

A program may call the MCP server too, with an API key as its Bearer token, by sending MCP’s JSON-RPC messages. This one calls get_org:
The answer is JSON, or one server-sent event whose data: line holds it. "method": "tools/list" lists the tools your key may use. An MCP client library does all of this for you: give it the address and the header Authorization: Bearer <your key>.