[SaaS only]{class="badge positive" title="Applies to Adobe Commerce as a Cloud Service and Adobe Commerce Optimizer projects only (Adobe-managed SaaS infrastructure)."}

AI coding tools for extensions

When migrating to Adobe Commerce as a Cloud Service, you can use the AI coding tools to convert existing Adobe Commerce PHP extensions to Adobe Developer App Builder extensions. You can also use these tools to create new App Builder extensions.

The AI coding tools provide the following benefits:

Prerequisites

Installation

The setup process prompts you with configuration options. For the setup location, choose “Current directory” to install the tools in your current workspace:

? Where would you like to setup the tools?
❯ Current directory
  New directory

When selecting the coding agent, Adobe recommends selecting Cursor for the best development experience:

? Which coding agent would you like to use?
❯ Cursor
  Copilot
  Gemini CLI
  Claude Code

When selecting the package manager, Adobe recommends using npm for consistency:

? Which package manager would you like to use?
❯ npm
  yarn

Post-installation configuration

Log in to the Adobe I/O CLI

After installing the Adobe I/O CLI, you must log in any time you want to use the MCP server.

aio auth login

To verify you are logged in, run the following command:

aio where

If you encounter issues, try logging out and logging back in:

aio auth logout
aio auth login

Cursor

What are the differences between Adobe Commerce PaaS and Adobe Commerce as a Cloud Service when configuring a webhook that activates an App Builder runtime action?

Copilot

Sample prompt

The following sample prompt creates an extension to send notifications when an order is placed.

Implement an Adobe Commerce SaaS extension that will send an ERP notification when a customer places an order. The ERP notification must be sent as a POST HTTP call to <ERP URL> with the following details in the request JSON body:

Order ID -> orderID
Order Total -> total
Customer Email ID -> emailID
Payment Type -> pType

Prompt commands

In addition to prompting, you can use the /search-commerce-docs command to search documentation in conversations with your agent. For example:

/search-commerce-docs "How do I subscribe to Commerce events?"

Best Practices

Adobe recommends following the following best practices when using the AI coding tools:

Checklist

Before starting any development session:

During development:

When starting new chat:

Workflow

When developing with the AI coding tools, start with sample code or scaffolded projects. This approach ensures you are building on a solid foundation rather than starting from nothing, while also optimizing your AI development workflow.

This also allows you to leverage Adobe’s templates and build upon proven patterns and architectures, while keeping established directory structures and conventions.

Consult the following resources to get started:

Why you should use these resources

Protocol

The following four-phase protocol is automatically enforced by the rules system. The tools should follow this protocol automatically when developing extensions:

Request implementation plans for complex development

For complex development involving multiple runtime actions, touchpoints, or integrations, explicitly request that the AI tools create a detailed implementation plan. When you see a high-level plan in Phase 2 that involves multiple components, ask for a detailed implementation plan to break it down into manageable tasks:

Create a detailed implementation plan for this complex development.

Complex Adobe Commerce extensions often involve:

Use MCP tools

The tooling defaults to MCP tools, but in certain circumstances it can use CLI commands instead. To ensure MCP tool usage, explicitly request them in your prompt.

If you see CLI commands being used and want to use MCP tools instead, use the following prompt:

Use only MCP tools and not CLI commands

CLI commands can be used for the following scenarios:

Development

Question unnecessary complexity created by the AI tools.

When unnecessary files are added (validator.js, transformer.js, sender.js) for simple read-only endpoints, use the following prompts:

Why do we need these files for a simple read-only endpoint?
Perform a root cause analysis before adding complexity
Verify if simpler solutions exist

Testing

Use the following best practices when testing:

Test each feature after implementation

After completing development of a feature in your implementation plan, test it immediately. Early testing prevents compound issues and makes debugging easier.

Test locally first

Always test locally first using the aio-app-dev tool. This provides immediate feedback and allows for faster iteration cycles, easier debugging, and no deployment overhead.

Deploy and test again

Once local testing is successful, deploy and test in the runtime environment. Runtime environments can have different behavior than local development.

Leverage the coding tools for testing support

Ask for help with testing. The tools can help with debugging, log analysis, and creating appropriate test data for your specific runtime actions.

Test runtime actions:

Help me test the customer-created runtime action running locally

Debug failures:

Why did the subscription-updated runtime action activation fail?

Check logs:

Help me check the logs for the last stock-monitoring runtime action invocation

Create test payloads:

Generate test data for this Commerce event

Create a test payload for the customer_save_after event

Find runtime endpoints:

What's the URL for this deployed action?

Handle authentication:

How do I authenticate with this external API?

Troubleshoot:

Help me debug why this action is returning 500 errors

Debugging

Stop and assess when things go wrong. If you encounter issues:

Deployment

Use the following best practices when deploying:

Deploy incrementally

Deploy only modified actions to speed up development. This approach reduces the risk of breaking existing functionality and provides quicker feedback on changes.

Runtime cleanup

After major changes, leverage the tools to clean up orphaned actions. Let the AI tooling handle the cleanup process systematically. It can efficiently identify orphaned actions, verify their status, and safely remove them without manual intervention.

Help me identify and clean up orphaned runtime actions

Request the AI tooling to list deployed actions and identify unused ones

List all deployed actions and identify which ones are no longer needed

Have the AI tools remove orphaned actions using appropriate commands

Remove the orphaned actions that are no longer part of the current implementation

Monitoring

Use the following best practices when monitoring your application:

Watch for context quality indicators

Track development velocity

Monitor cost efficiency

Track token usage patterns:

What to avoid

Avoid the following anti-patterns when using the AI coding tools: