For many apps, you can define a fixed schema when the application starts, and define it using GraphQL schema language. In some cases, it's useful to construct a schema programmatically. You can do this using the GraphQLSchema constructor.

When you are using the GraphQLSchema constructor to create a schema, instead of defining Query and Mutation types solely using schema language, you create them as separate object types.

For example, let's say we are building a simple API that lets you fetch user data for a few hardcoded users based on an id. Using buildSchema we could write a server with:

var express = require('express');
var { graphqlHTTP } = require('express-graphql');
var { buildSchema } = require('graphql');

var schema = buildSchema(`
  type User {
    id: String
    name: String
  }

  type Query {
    user(id: String): User
  }
`);

// Maps id to User object
var fakeDatabase = {
  a: {
    id: 'a',
    name: 'alice',
  },
  b: {
    id: 'b',
    name: 'bob',
  },
};

var root = {
  user: function ({ id }) {
    return fakeDatabase[id];
  },
};

var app = express();
app.use(
  '/graphql',
  graphqlHTTP({
    schema: schema,
    rootValue: root,
    graphiql: true,
  }),
);
app.listen(4000, () => {
  console.log('Running a GraphQL API server at localhost:4000/graphql');
});

We can implement this same API without using GraphQL schema language:

var express = require('express');
var { graphqlHTTP } = require('express-graphql');
var graphql = require('graphql');

// Maps id to User object
var fakeDatabase = {
  a: {
    id: 'a',
    name: 'alice',
  },
  b: {
    id: 'b',
    name: 'bob',
  },
};

// Define the User type
var userType = new graphql.GraphQLObjectType({
  name: 'User',
  fields: {
    id: { type: graphql.GraphQLString },
    name: { type: graphql.GraphQLString },
  },
});

// Define the Query type
var queryType = new graphql.GraphQLObjectType({
  name: 'Query',
  fields: {
    user: {
      type: userType,
      // `args` describes the arguments that the `user` query accepts
      args: {
        id: { type: graphql.GraphQLString },
      },
      resolve: function (_, { id }) {
        return fakeDatabase[id];
      },
    },
  },
});

var schema = new graphql.GraphQLSchema({ query: queryType });

var app = express();
app.use(
  '/graphql',
  graphqlHTTP({
    schema: schema,
    graphiql: true,
  }),
);
app.listen(4000, () => {
  console.log('Running a GraphQL API server at localhost:4000/graphql');
});

When we use this method of creating the API, the root level resolvers are implemented on the Query and Mutation types rather than on a root object.

This is particularly useful if you want to create a GraphQL schema automatically from something else, like a database schema. You might have a common format for something like creating and updating database records. This is also useful for implementing features like union types which don't map cleanly to ES6 classes and schema language.