Type safe GraphQL on Node.js


With everything moving to TypeScript and even Python + Ruby looking at adding some sort of types one could assume that strong typing is winning the battle against dynamic typing. Clear and strong types also makes you think twice about using JSON as a data transfer mechanism which is sort of schema / type less. Now I am not advocating to go back to XML+XSD days but I think the GraphQL specification provides decent type safety without adding any verbosity at all.

In fact GraphQL is pitched as - A query language for your API, and a server-side runtime for executing queries by using a type system you define for your data. So a type system forms the very core of GraphQL. On the flip side, the most popular implementation/s of GraphQL are on Node.js which inherently is not type friendly. In this post we will look at how to use TypeScript on both server and client side to work with GraphQL implementations for better type safety and hopefully fewer bugs.

Essentially what we do not want is to define our GraphQL schema in a giant string like -

// schema.js
const typeDefs = gql`
  type Book {
    title: String
    author: String

  type Query {
    books: [Book]

Instead we want to use the type capabilities already available in TypeScript. As usual we also want to use hapi as our main server framework with the endpoints secured by JWT. I am going to keep it simple and just say that typed GraphQL is available on TypeScript through the aptly named type-graphql package. With TypeScript setup our hapi "server.ts" combined with JWT auth & apollo-graphql looks like -

const init = async () => {

  const server = new Hapi.Server({
    port: 8080,
    host: 'localhost',
    routes: { cors: { origin: ['http://localhost:3000'] } }

  // JWT plugin setup
  await server.register(require('hapi-auth-jwt2'));
  server.auth.strategy('jwt', 'jwt', {
    key: SECRET,
    validate: validateToken,
    verifyOptions: { algorithms: ['HS256'] }

  // GraphQL setup
  const schema = await bootstrapSchema();
  const apolloServer = new ApolloServer({
    playground: true
  await apolloServer.applyMiddleware({
    app: server

  // Standard routes

  await server.start();
  console.log(`Server running on ${server.info.uri}`);

process.on('unhandledRejection', (err) => {


A simple GraphQL schema setup looks like -

import { Arg, Query, Resolver, buildSchema } from 'type-graphql';
import Greeting from '../models/greeting';

class GreetingResolver {
  @Query(returns => Greeting, { nullable: true })
  greet (@Arg('to', { nullable: false }) to: string): Greeting {
    return new Greeting(`Hello, ${to}!`);

export default async function bootstrapSchema () {
  return buildSchema({
    resolvers: [GreetingResolver],
    emitSchemaFile: true

type-graphql enables us to use standard TypeScript types to define our schema. The only catch is that we need to add a bit of "decorator magic". Decorators are really not stable on Node.js in the sense that it is an experimental feature that may change in future releases. So this something we need to keep in mind as EcmaScript standards evolve.

On the client side, while we cannot make the query itself type safe we can do something quite simple like -

class QueryBuilder {

  static greet = (to: string) => `{
    greet(to: "${to}") {


async executeGQLQuery<T>(queryName: string, queryBuilder: Function): Promise<T> {
  const token = await getToken();
  const response = await axios
      { query: queryBuilder() },
      { headers: { Authorization: `Bearer ${token}` } }

  return response.data.data[queryName];

// ...
// call the API endpoint from a redux-saga
const result = yield call(API.executeGQLQuery, 'greet', () => QueryBuilder.greet('World'));

This would ensure that we ony need to define the queries, while executing them needs only a single generic function definition. While GraphQL is already self-documenting, the advantage of this setup is that we have added type safety on top. It would be awesome if we had a library that could read a GraphQL schema and automatically generate the client side code in TypeScript.

The full working version of this setup is available on github.