Configuring CORS

Cross-Origin Resource Sharing (CORS) is an HTTP-header-based protocol that enables a server to dictate which origins can access its resources. Put another way, your server can specify which websites can tell a user's browser to talk to your server, and precisely which types of HTTP requests are allowed.

The startStandaloneServer function's CORS configuration is unalterable and enables any website on the internet to tell a user's browser to connect to your server. Depending on your use case, you might need to further customize your CORS behavior to ensure your server's security. To do so, you'll first need to swap to using expressMiddleware (or any other Apollo Server integration).

By default, websites running on domains that differ from your server's domain can't pass cookies with their requests. For details on enabling cross-origin cookie passing for authentication, see Passing credentials with CORS.

Why use CORS?

Most developers know about CORS because they run into the all-too-common CORS error. CORS errors usually occur when you set up an API call or try to get your separately hosted server and client to talk to each other. To better understand what CORS is and why we use it, we'll briefly go over some background context.

Internet users should always exercise caution when installing any new software on their devices. But when it comes to browsing the web, we navigate to different sites all the time, letting our browsers load content from those sites along the way. This comes with inherent risks.

As web developers, we don't want a user's browser to do anything fishy to our server while the user is visiting another website. Browser security mechanisms (e.g., CORS or SOP) can give developers peace of mind by enabling a website's server to specify which browser origins can request resources from that server.

The origin of a piece of web content consists of that content's domain, protocol, and port. The same-origin policy (SOP) is a security mechanism that restricts scripts on one origin from interacting with resources from another origin. This means that scripts on websites can interact with resources from the same origin without jumping through any extra hoops.

If two URLs differ in their domain, protocol, or port, then those URLs come from two different origins:

1
# Same origin
2
http://example.com:8080/ <==> http://example.com:8080/
3

4
# Different origin (difference in domain, protocol, and port)
5
http://example.com:8080/ =X= https://example1.com:8081/

However, as we all know, the internet is an exciting place full of resources that can make websites better (importing images, extra fonts, making API calls, and so on). Developers needed a new protocol to relax SOP and safely share resources across different origins.

Cross-Origin Resource Sharing is the mechanism that allows a web page to share resources across different origins. CORS provides an extra layer of protection by enabling servers and clients to define HTTP headers that specify which external clients' scripts can access their resources.

Note that both SOP and CORS are related to browser security. Neither prevents other types of software from requesting resources from your server.

Choosing CORS options for your project

When thinking about configuring CORS for your application, there are two main settings to consider:

• Which origins can access your server's resources
• Whether your server accepts user credentials (i.e., cookies) with requests

Specifying origins

CORS uses specific HTTP response headers as part of its protocol, including Access-Control-Allow-Origin. The Access-Control-Allow-Origin header (ACAO) enables a server to dictate which origins can use scripts to access that server's resources.

Depending on what you're building, the origins you specify in your CORS configuration might need to change when you're ready to deploy your application.

⚠️ Applications on private networks

If your browser is running your API on a private network (i.e., not on the public internet) and it relies on the privacy of that network for security, we strongly recommend specifying which origins can access your server's resources.

Specifically, the startStandaloneServer function's CORS behavior is not secure in this context. Instead, we recommend swapping to another Apollo Server integration to customize your server's CORS behavior by specifying origins.

If you don't, while your personal computer is on your private network, a script on any website could potentially make your browser talk to your private API. Some browsers, such as Chrome, have features under development to solve this problem. But in the meantime, all servers on private networks should always specify origins in their CORS configuration.

Federated subgraphs

If you're building a federated graph, we strongly recommend that all of your production subgraph servers disable CORS or limit your subgraph's origins to tools like the Apollo Studio Explorer. Most subgraphs should only allow communication from your gateways, and that communication doesn't involve a browser.

For more information, see Securing your subgraphs.

APIs that require cookies

If your API needs to accept cross-origin cookies with requests, you must specify origins in your CORS configuration. Otherwise, cross-origin cookies are automatically disabled. This is not a security vulnerability, but it does prevent your API from successfully providing cookies.

For examples, see Passing credentials with CORS.

APIs with known consumers

If you create an API on the public internet to serve resources to your own websites or apps, you might want to specify which origins can access your server's resources. Explicitly specifying origins can provide an extra level of security.

Public or embedded APIs

If you create a public API or an API to embed in websites you don't control yourself, you probably want to allow all origins to access your server's resources. You can use the startStandaloneServer's ACAO value (the wildcard *) in these situations, allowing requests from any origin.

None of the above

If your application doesn't fit into any of the above categories, startStandaloneServer's CORS behavior should suit your use case. You can always choose to swap to another Apollo Server integration later to customize your CORS configuration.

Configuring CORS options for Apollo Server

Apollo Server's standalone server (i.e., startStandaloneServer) serves the Access-Control-Allow-Origin HTTP header with the wildcard value (*). This allows scripts on any origin to make requests, without cookies, to the server and read its responses.

The startStandaloneServer function doesn't support configuring your server's CORS behavior. If you want to customize your server's CORS behavior (e.g., by specifying origins or passing cookies), swap to using expressMiddleware (or any other Apollo Server integration).

Below, we set up and customize the CORS behavior for our expressMiddleware function using the cors package:

1
import { ApolloServer } from '@apollo/server';
2
import { expressMiddleware } from '@apollo/server/express4';
3
import { ApolloServerPluginDrainHttpServer } from '@apollo/server/plugin/drainHttpServer';
4
import express from 'express';
5
import http from 'http';
6
import { json } from 'body-parser';
7
import { typeDefs, resolvers } from './schema';
8
import cors from 'cors';
9

10
const app = express();
11
const httpServer = http.createServer(app);
12
const server = new ApolloServer({
13
  typeDefs,
14
  resolvers,
15
  plugins: [ApolloServerPluginDrainHttpServer({ httpServer })],
16
});
17

18
await server.start();
19

20
app.use(
21
  '/graphql',
22
  cors<cors.CorsRequest>({ origin: ['https://www.your-app.example', 'https://studio.apollographql.com'] }),
23
  json(),
24
  expressMiddleware(server),
25
);
26

27
await new Promise<void>((resolve) => httpServer.listen({ port: 4000 }, resolve));
28
console.log(`🚀 Server ready at http://localhost:4000/graphql`);

1
import { ApolloServer } from '@apollo/server';
2
import { expressMiddleware } from '@apollo/server/express4';
3
import { ApolloServerPluginDrainHttpServer } from '@apollo/server/plugin/drainHttpServer';
4
import express from 'express';
5
import http from 'http';
6
import { json } from 'body-parser';
7
import { typeDefs, resolvers } from './schema';
8
import cors from 'cors';
9

10
const app = express();
11
const httpServer = http.createServer(app);
12
const server = new ApolloServer({
13
  typeDefs,
14
  resolvers,
15
  plugins: [ApolloServerPluginDrainHttpServer({ httpServer })],
16
});
17

18
await server.start();
19

20
app.use(
21
  '/graphql',
22
  cors({ origin: ['https://www.your-app.example', 'https://studio.apollographql.com'] }),
23
  json(),
24
  expressMiddleware(server),
25
);
26

27
await new Promise((resolve) => httpServer.listen({ port: 4000 }, resolve));
28
console.log(`🚀 Server ready at http://localhost:4000/graphql`);

Using the cors package directly, we can configure the Access-Control-Allow-Origin header using the origin option. The example above enables CORS requests from https://www.your-app.example, along with https://studio.apollographql.com.

Note that an origin doesn't include the path of a URL, meaning two URLs with different paths can still have the same origin. So when specifying an origin, don't include any paths or trailing slashes (e.g., use http://example.com, not http://example.com/).

If you're using another integration of Apollo Server, you can add and configure CORS for your server using your framework's standard functionality.

You can also choose to omit CORS middleware entirely to disable cross-origin requests. This is recommended for subgraphs in a federated graph.

Passing credentials with CORS

If your server requires requests to include a user's credentials (e.g., cookies), you need to modify your CORS configuration to tell the browser those credentials are allowed.

You can enable credentials with CORS by setting the Access-Control-Allow-Credentials HTTP header to true.

To enable your browser to pass credentials using expressMiddleware and the cors npm package, you can specify origins and set credentials to true:

1
app.use(
2
  '/graphql',
3
  cors<cors.CorsRequest>({
4
    origin: yourOrigin,
5
    credentials: true,
6
  }),
7
  json(),
8
  expressMiddleware(server),
9
);

1
app.use(
2
  '/graphql',
3
  cors({
4
    origin: yourOrigin,
5
    credentials: true,
6
  }),
7
  json(),
8
  expressMiddleware(server),
9
);

For examples of sending cookies and authorization headers from Apollo Client, see Authentication. For more guidance on adding authentication logic to Apollo Server, see Authentication and authorization.

Preventing Cross-Site Request Forgery (CSRF)

Your server's CORS policy enables you to control which websites can talk to your server. In most cases, the browser checks your server's CORS policy by sending a preflight request before sending the actual operation. This is a separate HTTP request. Unlike most HTTP requests (which use the GET or POST method), this request uses a method called OPTIONS. The browser sends an Origin header, along with some other headers that start with Access-Control-. These headers describe the kind of request that the potentially untrusted JavaScript wants to make. Your server returns a response with Access-Control-* headers describing its policies (as described above), and the browser uses that response to decide whether it's OK to send the real request. Processing the OPTIONS preflight request never actually executes GraphQL operations.

However, in some circumstances, the browser will not send this preflight request. If the request is considered "simple", then the browser just sends the request without sending a preflight first. Your server's response can still contain Access-Control-* headers, and if they indicate that the origin that sent the request shouldn't be able to access the site, the browser will hide your server's response from the problematic JavaScript code.

Unfortunately, this means that your server might execute GraphQL operations from "simple" requests sent by sites that shouldn't be allowed to communicate with your server. And these requests can even contain cookies! Although the browser will hide your server's response data from the malicious code, that might not be sufficient. If running the operation has side effects, then the attacker might not care if it can read the response or not, as long as it can use an unsuspecting user's browser (and cookies!) to trigger those side effects. Even with a read-only query, the malicious code might be able to figure out something about the response based entirely on how long the query takes to execute.

Attacks that use simple requests for their side effects are called "cross-site request forgery" attacks, or CSRF. Attacks that measure the timing of simple requests are called "cross-site search" attacks, or XS-Search.

To avoid CSRF and XS-Search attacks, GraphQL servers should refuse to execute any operation coming from a browser that has not "preflighted" that operation. There's no reliable way to detect whether a request came from a browser, so GraphQL servers should not execute any operation in a "simple request".

The most important rule for whether or not a request is "simple" is whether it tries to set arbitrary HTTP request headers. Any request that sets the Content-Type header to application/json (or anything other than a list of three particular values) cannot be a simple request, and thus it must be preflighted. Because all POST requests recognized by Apollo Server must contain a Content-Type header specifying application/json, we can be confident that they are not simple requests and that if they come from a browser, they have been preflighted.

However, Apollo Server also handles GET requests. GET requests do not require a Content-Type header, so they can potentially be simple requests. So how can we ensure that we only execute GET requests that are not simple requests? If we require the request to include an HTTP header that is never set automatically by the browser, then that is sufficient: requests that set HTTP headers other than the handful defined in the spec must be preflighted.

By default, Apollo Server 4 has a CSRF prevention feature enabled. This means your server only executes GraphQL operations if at least one of the following conditions is true:

• The incoming request includes a Content-Type header that specifies a type other than text/plain, application/x-www-form-urlencoded, or multipart/form-data. Notably, a Content-Type of application/json (including any suffix like application/json; charset=utf-8) is sufficient. This means that all POST requests (which must use Content-Type: application/json) will be executed. Additionally, all versions of Apollo Client Web that support GET requests do include Content-Type: application/json headers, so any request from Apollo Client Web (POST or GET) will be executed.
• There is a non-empty X-Apollo-Operation-Name header. This header is sent with all operations (POST or GET) by Apollo iOS (v0.13.0+) and Apollo Kotlin (all versions, including its former name "Apollo Android"), so any request from Apollo iOS or Apollo Kotlin will be executed.
• There is a non-empty Apollo-Require-Preflight header.

Note that all HTTP header names are case-insensitive.

HTTP requests that satisfy none of the conditions above will be rejected with a 400 status code and a message clearly explaining which headers need to be added in order to make the request succeed.

This feature should have no impact on legitimate use of your graph except in these two cases:

• you have clients that send GET requests and are not Apollo Client Web, Apollo iOS, or Apollo Kotlin
• you have enabled file uploads with graphql-upload

If either of these apply to you, you should configure the relevant clients to send a non-empty Apollo-Require-Preflight header along with all requests.

You can also configure the set of headers that allow execution. For example, if you use a GraphQL client that performs GET requests without sending Content-Type, X-Apollo-Operation-Name, or Apollo-Require-Preflight headers, but it does send a Some-Special-Header header, you can pass csrfPrevention: { requestHeaders: ['Some-Special-Header'] } to new ApolloServer(). This option replaces checking for the two headers X-Apollo-Operation-Name and Apollo-Require-Preflight in the CSRF prevention logic. The check for Content-Type remains the same.

Note that Apollo Server does not permit executing mutations in GET requests (just queries). As long as you ensure that only mutations can have side effects, you are somewhat protected from the "side effects" aspect of CSRFs even without enabling CSRF protection. However, you would still be vulnerable to XS-Search timing attacks.

graphql-upload

The third-party graphql-upload package has a known CSRF vulnerability.

The graphql-upload package adds a special middleware that parses POST requests with a Content-Type of multipart/form-data. This is one of the three special Content-Types that can be set on simple requests, enabling your server to process mutations sent in simple requests.

Apollo Server 4's default CSRF prevention feature blocks those attempting to use the graphql-upload package. If you use graphql-upload you should keep the CSRF prevention feature enabled, and configure your upload clients to send a non-empty Apollo-Require-Preflight header.

For example, if you use the apollo-upload-client package with Apollo Client Web, pass {headers: {'Apollo-Require-Preflight': 'true'}} to createUploadLink.

For more information about file upload best practices, see our blog post.