StoreFront Agent Toolchain
Overview
The UltraCart StoreFront Agent Toolchain lets an AI coding agent that you already use, such as Claude Code, Codex or Gemini CLI, build and edit your UltraCart StoreFront. You describe the change in plain language. The agent carries it out with UltraCart's own commands, shows you a preview, and stops to ask you before anything risky happens.
The toolchain has three parts:
- The
uc-storefrontcommand line tool (CLI). It is the npm package@ultracart/storefront-cli. It reads, validates and writes StoreFront content through guarded commands. - A packaged catalog of StoreFront elements. It tells the agent which settings each element accepts, so the agent works from real contracts instead of guessing.
- An agent skill named
build-ultracart-storefronts. It teaches the agent how to use the CLI safely: which reference to read before each kind of change, how to preview, and when to stop and ask you.
It is not an MCP (Model Context Protocol) server. Your agent runs uc-storefront commands in a terminal and reads their output, the same way it runs any other command line tool.
Status: public beta
The toolchain is in public beta. The CLI is published on public npm under the beta tag, and you can install it today. Expect changes between releases.
Installing the CLI does not turn on access to your store. UltraCart enables merchant accounts for the beta. An account that is not enabled yet gets a beta_access_required response when the CLI first talks to UltraCart. See Connect your store for what that means and what to do.
What it costs
The toolchain is free to use. You bring your own AI agent and pay for that agent's subscription with its provider.
Some UltraCart features the agent can switch on carry their own charges, and the toolchain does not change that. For example, turning on another storefront language is billed for machine translation, screen recording is billed once its trial ends, postcards are printed and mailed at the account's cost, SMS steps are billed per segment, and warehouse queries run against your own Google BigQuery project. The capability pages note these where they apply. The agent skill tells the agent to leave these decisions to you.
What you can do with it
With the toolchain, your agent can work on most of a StoreFront through one guarded loop: read what is there, change it locally, validate it, preview it, then write it.
- Build and restyle pages, headers, footers, checkout and My Account.
- Change item content, attributes, SEO fields and prices.
- Build upsell paths and offers, switched off until you approve them.
- Build campaigns, flows, segments, emails and postcards in StoreFront Communications, and preview them without sending.
- Read your BigQuery data warehouse, screen recordings and server render logs.
- Manage languages, redirects, 404s, experiments, blog posts, surveys, and files.
What it can do has one page per area, with example requests, the commands involved, what is out of scope, and what needs your approval.
Who it is for
- Merchants who want to describe a StoreFront change and have an agent make it, without learning the Visual Builder or Velocity in depth. You stay in control: the agent previews first, and you approve risky changes in the UltraCart admin.
- Developers and agencies who build StoreFronts for one or more merchants and want a repeatable, scriptable workflow. Every command accepts
--jsonfor machine-readable output, and setup can run without prompts.
You need an UltraCart account. To try it with less risk, start on a test storefront or an inactive theme. Some changes, such as item content, menus and site settings, are shared by every theme and go live as soon as they are written (see Install and set up).
How it differs from an assistant built into an admin
The toolchain does not add a chat window to the UltraCart admin. It works from outside, through the agent you choose:
- It runs in your agent, on your computer or a host you control. You pick the agent and its model, and you keep the agent's normal file, terminal and project tools.
- Work happens in a project folder. The agent pulls content into local files, edits and validates them there, and keeps originals so a change can be undone.
- Writes go through guarded commands, not free-form API calls. The CLI reserves real widget IDs, validates content, checks that nothing changed since the agent read it, and refuses a live change without an explicit acknowledgement.
- You approve risky changes yourself, in the UltraCart admin. The agent cannot approve its own requests. See Safety and human approval.
Related documentation
- Install and set up: install the CLI and the agent skill
- Connect your store: sign in and bind a workspace to a merchant
- Safety and human approval: what the guardrails do, and what needs your approval
- Troubleshooting: common setup problems and their fixes
- StoreFront Visual Builder: the merchant guide to the editor whose content the agent changes