Kotlin Multiplatform + Pizza = ❤️

Step 11: Add networking layer to shared KMM module with Ktor

Now that we have a backend for the recipes we’ll use KMM power to implement some shared layers, such as networking, local app storage and even viewmodels.

Let’s start with networking, where we will use all the functionality from our backend.

First, add Ktor Client dependencies to our project. Yes, we’ll be using the same Ktor we used for our backend, but it’s Client functionality this time.

Add a Common object with necessary dependencies to your Versions.kt:

object Common {
   const val KTOR_CLIENT_CORE = "io.ktor:ktor-client-core:$KTOR_VERSION"
   const val KTOR_CLIENT_JSON = "io.ktor:ktor-client-json:$KTOR_VERSION"
   const val KTOR_LOGGING = "io.ktor:ktor-client-logging:$KTOR_VERSION"
   const val KTOR_CLIENT_SERIALIZATION = "io.ktor:ktor-client-serialization:$KTOR_VERSION"

Alongside with the client dependency and additional features like logging and serialization we add engine dependencies for our platform sides:

object Android {
   const val KTOR_CLIENT = "io.ktor:ktor-client-android:$KTOR_VERSION"
   const val KTOR_OKHTTP = "io.ktor:ktor-client-okhttp:$KTOR_VERSION"

object iOS {
   const val KTOR_CLIENT = "io.ktor:ktor-client-ios:$KTOR_VERSION"

Now implement these dependencies in build.gradle.kts in your shared module (not backend!). Modify the corresponding sourceSets as follows:

val commonMain by getting {
   dependencies {
       // Ktor

val androidMain by getting {
   dependencies {

val iosMain by getting {
   dependencies {

Also add serialization plugin to the shared module plugins, so that we can deserialize data from network responses into data classes:

plugins {
   kotlin("plugin.serialization") version Versions.KOTLIN_VERSION

Sync your project files. Now, let’s create our Ktor client. First, in the shared module in commonMain create a new package api and add a KtorApi interface:

create KtorApi interface

Here we declare the Ktor client and the methods which we’ll use for all api calls.

interface KtorApi {
   val client: HttpClient
   fun HttpRequestBuilder.apiUrl(path: String)
   fun HttpRequestBuilder.json()

With apiUrl we’ll set the request path and with json we’ll tell the client to receive data in form of json.

Then we add a KtorApiImpl.kt inext to KtorApi.kt and write an implementation:

class KtorApiImpl() : KtorApi {

   val prodUrl = "" [1]

       override val client = HttpClient {
       install(JsonFeature) { [2]
           serializer = KotlinxSerializer()

   override fun HttpRequestBuilder.apiUrl(path: String) {
       url {
           takeFrom(prodUrl) [3]
           encodedPath = path

   override fun HttpRequestBuilder.json() {
       contentType(ContentType.Application.Json) [4]

[1] For now we hardcoded the backend urls, later we’ll see how to manage them in build configuration
[2] We’ll need to serialize and deserialize our objects to receive and send them to our backend, therefore we use kotlinx serializer feature, that we imported earlier
[3] Specify the url that should be used to fetch data: it consists of the base production url and the following endpoint path
[4] Specify the content type for all requests, here we are using json

Now let’s create our actual RecipesApi. Your commonMain should look like this by now:

create KtorApi interface

RecipeApi will extend the KtorApi and contain the recipe endpoint:

class RecipesApi(private val ktorApi: KtorApi) : KtorApi by ktorApi {
   companion object {
       const val RECIPES_BASE_URL = "recipes"

Also add all the necessary requests for our app:

suspend fun getPizza(): String {
   return client.get {
       apiUrl("pizza") [1]

suspend fun getRecipes(): List<RecipeResponse> {
   return client.get { [2]

suspend fun getRecipe(id: Long): RecipeResponse {
   return client.get {

suspend fun postRecipe(recipe: Recipe): Long [3] {
   return {
       json() [4]

[1] For our test endpoint we don’t use the recipes url: instead we call the pizza url and should get a String in return
[2] Use the GET function to receive data from backend as a List of RecipeResponses
[3] When we post a recipe we get an ID of type Long in return
[4] Don’t forget to specify the content type for your post request

As you can see, we just mirrored the requests that we already have in our backend api. We also have an option to upload pictures, but we’ll see how to do it after we’re done with basic text requests.

You may be wondering, why Recipe and RecipeRepsonse are unresolved. Well, we’ve created an entity for our model, but it was in another module, the backend module.

To solve this problem, let’s copy that Entity.kt file and put it into a model folder in our shared module. Now you can resolve the references in RecipeApi.

Let’s remove the model package from our backend module and implement the shared project there instead.

To do so, in shared build.gradle.kts first add jvm() to your targets:

kotlin {

. . .

Now, implement the shared project in the backend build.gradle.kts under dependencies:


Sync your project files and fix the Exposed entities in your backend module to import corresponding classes from kmpizza.model package.

Now we have our model only in one location, and it’s extremely convenient, isn’t it?

Finally, let’s create a remote data source, which we can later use alongside a local source to get appropriate data.

Create RecipeRemoteSource class in shared/commonMain/remote.

create KtorApi interface

This class will receive recipesApi as a parameter and use it to fetch recipes with getRecipes from our RecipesApi, as well as add a recipe with postRecipe.

class RecipeRemoteSource(
   private val recipesApi: RecipesApi
) {

   suspend fun getRecipes() = recipesApi.getRecipes()
   suspend fun postRecipe(recipe: Recipe) = recipesApi.postRecipe(recipe)

Easy as that, with the help of Ktor we’ve just added a networking layer to our shared KMM module. This will be the only location where we implement our api requests. You’ll see how to use them on platform sides in the following steps. However, we’re not done with the shared part. In the next step we’ll look into sharing a viewmodel layer.