Creating a new app

Version: 1.0.0



In this tutorial, you'll learn how to integrate new services into by creating a app on Deco Hub. We'll use an external API (RandomData) as an example, but you can use apps for several use cases, like:

  • Review providers (e.g: PowerReviews, Opiniões Verificadas).
  • External CMSes (e.g: Wordpress, Sanity)
  • UI widgets (e.g: SizeBay, chatbots)

Forking the deco-cx/apps repository

The first step to develop an app is to fork the deco-cx/apps. You can do it inside Github by accessing the repo link.

Forking deco-cx apps repository

After forking it, you'll have your own version of deco-cx/apps, usually hosted on your personal organization ({yourusername}/apps). This repository will be used to develop the app and, after finished, you'll submit a Pull Request to the original repository.

If you have questions about this flow, read about contributing to open-source.


In the root folder of the repo, you'll see a lot of folders. Each one of those represents an app. It's recommended that you take a look into those examples so you can better understand.

List of apps in deco-cx/apps

The example use case

To better illustrate the app's capabilities and also demonstrate how to convert ecommerce data, we will build an app called random-products and it'll simulate an external search provider, but getting random data from Random Data API. We will use the endpoint which returns a list of beers, and will complement with mocked information (like the product's image and price).

We'll create a productList.ts loader which can be used in's Product Shelves and other components that require Product[] | null.

Creating an app

It's important to first understand what you want to accomplish. If you want to extend an existing application, you don't need to create a new app, just create/update files inside the app that already exists. Since we want to create an integration with a new service, let's create an app:

  1. Inside the apps root folder, run deno task new random-product.
  2. Open the code in your IDE and verify that a random-product folder was created.


The mod.ts is the entrypoint of the application, and it defines the configuration type for the app. The most common usecase for this is to declare metadata that is important to all of the app's capabilities, like API Keys, account names, language. You can check the example in typesense/mod.ts. These values will be filled by the end user whenever they want to install the app.

Since our API (Random Data) doesn't require any specific authentication key, let's include only one prop for the root endpoint (in this case, it needs to be filled with

export type Props = {
  /** @description Use if not specified*/
  rootEndpoint: string;

Tip: You can use any of the Utility Types that you already know.

It's also in this file that you can make clients/services available to the app's loaders and actions, in order to simplify REST/GraphQL calls. Let's check how to use this feature.

Creating the HTTP client

Most of the integrations will require making requests to an external API via REST or GraphQL. In order to provide a better Developer Experience, we provide a createHTTPClient and createGraphqlClient that can be used to call external endpoints with type-safety. You can see an example of this in wake/mod.ts. automatically generates typing definitions for OpenAPI specifications. Just drop the JSON export for the api in *.openapi.json file and run deno task start. You can see an example in the wake app.

Let's model the API we're targeting in a api-typings.d.ts file in the app. To quickly generate the response's type, use QuickType. The end result looks like this:

export interface Beer {
  id: number;
  uid: string;
  brand: string;
  name: string;
  style: string;
  hop: string;
  yeast: string;
  malts: string;
  ibu: string;
  alcohol: string;
  blg: string;

export interface RandomDataApi {
   * Random beers
  "GET /api/v2/beers": {
    response: Beer[];
    searchParams: {
      size?: number;

Now, let's return to our mod.ts and properly create the client using the typing definition we've just created. After adjusting the types, the file will look like this:

import type { App, AppContext as AC } from "deco/mod.ts";
import manifest, { Manifest } from "./manifest.gen.ts";
import { createHttpClient } from "../utils/http.ts";
import { fetchSafe } from "../utils/fetch.ts";
import type { RandomDataApi } from "./api-typings.d.ts";

export type Props = {
  /** @description Use if not specified */
  rootEndpoint: string;

 * @title random-product
export default function App(
  { rootEndpoint }: Props,
) {
  const randomApi = createHttpClient<RandomDataApi>({
    base: `${rootEndpoint}`,
    headers: new Headers({
      "content-type": "application/json",
    fetcher: fetchSafe,

  const state = {

  const app: App<Manifest, typeof state> = {
  return app;

export type AppContext = AC<ReturnType<typeof App>>;

You can check other client example in vtex/mod.ts, using both HTTP and GraphQL.

Creating productList loader

Since our goal is to return product data that can be used in a Product Shelf UI, let's create a Loader that does just that. In this loader, we will:

  1. Declare the props that users will fill in deco's admin.
  2. Call the Random Data API using the randomApi client.
  3. Transform the data returned from the API into deco's Product type.

The 3rd step (transforming the data) is quite iterative, so it's better to have the loader working first and then work on the transformation. So, follow these steps:

  1. Create a loaders folder inside of the random-product app folder.
  2. Create a productList.ts file.
  3. Paste the following boilerplate:
import type { Product } from "../../commerce/types.ts";
import { AppContext } from "../mod.ts";

export interface Props {
  /** @description total number of items to display */
  count: number;

 * @title Random Products
const loader = async (
  props: Props,
  _req: Request,
  ctx: AppContext,
): Promise<Product[] | null> => {
  const { randomApi } = ctx;

  const count = props.count ?? 12;

  const beers = await randomApi["GET /api/v2/beers"]({ size: count }).then((
  ) => r.json());

  console.log({ beers });

  const products = await Promise.resolve([]);

  return products ?? [];

export default loader;

Now that the loader is created (returning only an empty array for now), let's connect it with a deco site and see it working in the Admin.

Testing our app locally

In order to test our loader, we need to install the app in a deco site. And, to make the random-product app installable, we first need to include it in Deco Hub (which itself is an app) and "build" our application. To do that:

  1. Create a random-product.ts file inside decohub/apps.
  2. Paste the following line:

export { default } from "../../random-product/mod.ts";

  1. Run deno task start in the apps root folder (not apps/deco-hub/apps, but apps.)

After this, the manifests will be generated and the apps will be ready. If anything is not right, the process will let you know.

Now, to test our app, let's use a deco site we have available and link it locally so we can check that everything is working fine. I'll use storefront but you can use any other site that have decohub.ts installed. If you don't have a site yet, create a new one and select storefront-vtex as a template.

  1. Place the site's repository in the same folder as the apps repo you've been working with.
  2. Open the site's code and then open the import_map.json file.
  3. Replace the value of the "apps" entry to "../apps/.
  4. Run deno task start in the site's root folder.

If everything is ok, the site should be running and accessible via http://localhost:8000.

Let's go to deco's admin for this site to finish our setup.

  1. Access http://localhost:8000.
  2. Press . on your keyboard to be redirected to the CMS.
  3. Click on Apps to go to the Apps page in Admin.
  4. In the top right, change the environment to http://localhost:8000.
  5. Search for random-product.ts. You should see a row representing the app that was just created. Refresh the page otherwise.
  6. Click on the + button to start instaling that app.
  7. Fill in the Root Endpoint field.
  8. Click on Save and give it a meaningful name (e.g: random-product-app).
  9. After having saved the block, click on Publish.

Now, the app is installed in that site. This means that our productList.ts loader should be available to use. Let's check:

  1. Go back to the Blocks page.
  2. Select the Loaders tab.
  3. Search for productList and select the one that is from random-product.
  4. Fill the Count with a number like 4.
  5. You should see an empty response (because the loader is returning []) but the list of beers should be logged in the server's console.
Running a loader in Console of the running loader

Ok, now the loader is running and logging data that comes from the Random Data API.

Transforming the data

Our objective with this loader is to power UI components that display products (like the Product Shelf) and, in order to do this, we need to transform our data into the expected type for that Sections. uses the's data type in its components so that the community can use the same code to connect with different ecommerce platforms. Because of that, we need to transform the data inside our loader to match the same type.


Let's create a utils/transform.ts file in our app which will hold the transforming function from the input type (Beer) to the output type (Product). Let's start with this snippet:

import type { Product } from "../../commerce/types.ts";
import { DEFAULT_IMAGE } from "../../commerce/utils/constants.ts";
import { Beer } from "../api-typings.d.ts";

export const toProduct = (
  beer: Beer,
  url: URL,
): Product => {
  // @ts-expect-error: Still a work in progress
  return {};

This tutorial only includes the Product List loader, but if you're integrating a completely new commerce platform you'll need to also transform data related to SEO and Search Filters.

Let's call this function inside of our productList loader, mapping our data over this function:

import { toProduct } from "../utils/transform.ts"
// ...

  const products = => toProduct(beer, new URL(req.url)));

  return products ?? [];

The toProduct function is not doing anything right now, but it's connected to the loader. From now on, we need to understand the data and create the mappings. This is a very iterative step and I recommend that you do it already testing with a UI section. Let's preview a ProductShelf in deco's admin connected with our loader and see where it falls off.

Error Cannot read properties of undefined (reading 'url') previewing a Product Shelf

You should see some errors because the section access some properties that we are not passing yet (like the one above), but seeing this can be helpful as it leads to what needs to be resolved in the transform function.

Since we're using a mock API that doesn't provide a lot of data, we'll have to hardcode some values. This is the final mapping function we've created:

export const toProduct = (
  beer: Beer,
  url: URL,
): Product => {
  return {
    "@type": "Product",
    productID: beer.uid,
    url: `${url.origin}/beers/${beer.uid}`,
    sku: `${}`,
    brand: { "@type": "Brand", name: beer.brand },
    additionalProperty: [{
      "@type": "PropertyValue",
      name: "hop",
      value: beer.hop,
    image: [DEFAULT_IMAGE],
    offers: {
      "@type": "AggregateOffer",
      priceCurrency: "BRL",
      highPrice: 100,
      lowPrice: 100,
      offerCount: 1,
      offers: [{
        "@type": "Offer",
        price: 100,
        availability: "",
        inventoryLevel: { value: 10 },
        priceSpecification: [{
          "@type": "UnitPriceSpecification",
          priceType: "",
          price: 100,

After saving the code and checking the ProductShelf again, we should see the following:


And that's it! The Random Product productList loader is working and can already be integrated into deco's UI sections.

We are using sample data that doesn't represent what's usually returned from search and ecommerce providers. In a real scenario, it's important to correcly map the data to's data types.


Now that the app is already and included in deco hub, create a Pull Request and send us at (#deco-prs channel).

Was this page helpful?
Can't find what you're looking for? Spot an error in the documentation? Get in touch with us on our Community Forum
Continue reading