Overview of building with AI

Ask about this Page
Copy for LLM
View as Markdown

The commercetools AI plugin gives your coding agent accurate and up-to-date knowledge of commercetools APIs, and proven implementation patterns.

Building on commercetools with AI

commercetools provides you with an official AI plugin. It's a single package that gives your coding agent the same current, trustworthy knowledge that your most senior commercetools architect keeps in their head.

Two things make the plugin work, and they're better together:

  • A curated set of skills, which are tested implementation playbooks that the agent loads as context.
  • commercetools Knowledge MCP, a live connection to current docs, API schemas, and query validation.

With the plugin, you get an agent that builds the first time correctly. It uses current API data instead of outdated training data, with correct field names, the right OAuth scopes, and awareness of the caveats that matter in production.

Install the plugin

In any Claude Code session:

/plugin marketplace add commercetools/commercetools-skills
/plugin install commercetools@commercetools
Reload plugins

If you've updated the plugin or installed it in another window and need the current session to pick up the latest version:

/reload-plugins
Claude Desktop
Customize -> Personal plugins -> Create plugin -> Add marketplace -> add commercetools/commercetools-ai-plugins

then click on the plugin and click "Install"

Why the plugin gets you to value faster

commercetools is powerful because it doesn't force your hand as it provides you with APIs and lets you assemble exactly the experience you want. However, the cost of that freedom might be the learning curve. Building with the AI plugin flattens that curve, and here's why:

Without pluginWith plugin
Read docs, prototype, get an API shape wrong, re-read, correct the AI a few times, debug a method that doesn't exist.The agent knows the idiomatic pattern and writes it correctly on the first pass.
The AI invents a feature or borrows an API from some other commerce vendor.The agent retrieves the Knowledge MCP and checks it against the live schema.
Every team designs from scratch, so no two storefronts look alike.Every team starts from the same proven foundation.
Weeks to a working reference implementation.A running storefront in a single working session.

If you're running the business or the engineering org, the takeaway is simple: the time and cost of standing up a commercetools storefront decreases, and the quality floor goes up. Every developer's assistant now produces specialist-level output. If you're the one writing the code, you spend less time fighting boilerplate and bad autocomplete. The rest of your time goes to the parts that are actually yours: the brand, the merchandising, and the experience nobody else can copy.

Skills and MCP: when each is used

These are two different tools, and the whole point is to use both. A skill is the senior engineer's instinct for "here's how we always do this", loaded whenever a task matches. The MCP provides access to the exact current spec. One gives you consistency, the other gives you currency, and together, the agent is opinionated and accurate at the same time.

SkillsMCP
What it isPersistent context the agent loads when a task matches.A live connection the agent calls when it needs a current fact.
Think of it as"Here's how we always build this.""Let me check the exact specs right now."
Where it livesIn the repo, as SKILL.md files plus a references/ folder.On a server endpoint, queried on demand.
What it gives youConsistency across teams and projects.Currency, always live, no staleness.
ExampleThe right way to build a cart-to-checkout flow.Validating a GraphQL query or confirming a field's shape.

What's in the plugin

The plugin provides you with two MCP servers:

  • commercetools-knowledge: it provides live documentation search, GraphQL and OpenAPI schema lookup, query validation, and developer best practices. Public endpoint, no API key, on by default.
  • commerce-mcp (optional): it lets the agent talk directly to your own commercetools project's APIs through function calling. Opt-in, and it needs your API client credentials before it'll do anything.

The plugin also comes with three skills:

  • commercetools-platform: the foundation everything else sits on. TypeScript SDK setup, ClientBuilder authentication, the project data model (Products, Customers, Orders, Types, Channels, and Stores), GraphQL versus REST, query predicates, optimistic concurrency, rate limits, and observability.
  • commercetools-storefront: the shared base for B2C and B2B storefronts in Next.js. BFF architecture, cart, checkout, product listing, product detail, search, facets, plus optional modules like bundles, promotions, impersonation, quotes, and approval workflows.
  • commercetools-checkout: Checkout pre-built solution. Payment-only mode, full hosted checkout, express payments, and PSP connector setup for Stripe, Adyen, and Mollie.

And on top of that, you get these slash commands:

  • /commercetools-nextjs-setup-project that scaffolds a new storefront.
  • /deploy-vercel and /deploy-netlify that handle deployment without accidentally shipping the wrong credentials.

What the plugin fixes, and what it does not

The plugin saves you from wasting time on:

  • Hallucinated SDK calls and invented features
  • Outdated API patterns dragged in from training data
  • The slow loop where you correct the AI three times before it gets the shape right
  • Architectural drift across teams
  • The dangerous stuff: credentials leaking into the browser, a client instantiated on every request, a missing scope guard

However, the plugin does not:

  • Write your requirements. The agent builds what you ask for, and builds it well, but it doesn't decide what to ask for.
  • Own your business logic and merchandising rules. Those stay yours.
  • Replace review and testing. The agent writes the code; your team owns it.
  • Harden anything for production beyond what the skills prescribe, so your own QA, security review, and load testing still matter.

What you need to use the plugin

On the commercetools side, you need:

On your machine, you need:

  • Node.js 22
  • Git
  • A supported coding agent. The plugin is tested with Claude Code, Cursor, VS Code Copilot, and OpenAI Codex, but any agent that reads SKILL.md and works with MCP can use it.

What to keep in mind

This is an official commercetools tooling, but treat the code your agent generates as a strong starting point, not a certified production artifact. Review it, test it, own it. The skills are also opinionated about stack and architecture, for more information, see Tech stack and architectural choices behind skills. That opinion is the reason they're fast. If your stack differs significantly, you can see where the friction shows up. Agentic loops burn tokens, and the MCP servers have rate limits. Build deliberately rather than sending one large, open-ended prompt.
When you're ready, go to Build a storefront from scratch to see it run. Read the Architecture and stack assumptions if you're evaluating the stack. Review Security and best practices before you go anywhere near production.