Home

Awesome

chrome-remote-interface

CI status

Chrome Debugging Protocol interface that helps to instrument Chrome (or any other suitable implementation) by providing a simple abstraction of commands and notifications using a straightforward JavaScript API.

This module is one of the many third-party protocol clients.

Sample API usage

The following snippet loads https://github.com and dumps every request made:

const CDP = require('chrome-remote-interface');

async function example() {
    let client;
    try {
        // connect to endpoint
        client = await CDP();
        // extract domains
        const {Network, Page} = client;
        // setup handlers
        Network.requestWillBeSent((params) => {
            console.log(params.request.url);
        });
        // enable events then start!
        await Network.enable();
        await Page.enable();
        await Page.navigate({url: 'https://github.com'});
        await Page.loadEventFired();
    } catch (err) {
        console.error(err);
    } finally {
        if (client) {
            await client.close();
        }
    }
}

example();

Find more examples in the wiki. You may also want to take a look at the FAQ.

Installation

npm install chrome-remote-interface

Install globally (-g) to just use the bundled client.

Implementations

This module should work with every application implementing the Chrome Debugging Protocol. In particular, it has been tested against the following implementations:

ImplementationProtocol versionProtocolListNewActivateCloseVersion
Chrometip-of-treeyes¹yesyesyesyesyes
Operatip-of-treeyesyesyesyesyesyes
Node.js (v6.3.0+)nodeyesnonononoyes
Safari (iOS)partialnoyesnononono
Edgepartialyesyesnononoyes
Firefox (Nightly)partialyesyesnoyesyesyes

¹ Not available on Chrome for Android, hence a local version of the protocol must be used.

The meaning of target varies according to the implementation, for example, each Chrome tab represents a target whereas for Node.js a target is the currently inspected script.

Setup

An instance of either Chrome itself or another implementation needs to be running on a known port in order to use this module (defaults to localhost:9222).

Chrome/Chromium

Desktop

Start Chrome with the --remote-debugging-port option, for example:

google-chrome --remote-debugging-port=9222
Headless

Since version 59, additionally use the --headless option, for example:

google-chrome --headless --remote-debugging-port=9222

Android

Plug the device and make sure to authorize the connection from the device itself. Then enable the port forwarding, for example:

adb -d forward tcp:9222 localabstract:chrome_devtools_remote

After that you should be able to use http://127.0.0.1:9222 as usual, but note that in Android, Chrome does not have its own protocol available, so a local version must be used. See here for more information.

WebView

In order to be inspectable, a WebView must be configured for debugging and the corresponding process ID must be known. There are several ways to obtain it, for example:

adb shell grep -a webview_devtools_remote /proc/net/unix

Finally, port forwarding can be enabled as follows:

adb forward tcp:9222 localabstract:webview_devtools_remote_<pid>

Opera

Start Opera with the --remote-debugging-port option, for example:

opera --remote-debugging-port=9222

Node.js

Start Node.js with the --inspect option, for example:

node --inspect=9222 script.js

Safari (iOS)

Install and run the iOS WebKit Debug Proxy. Then use it with the local option set to true to use the local version of the protocol or pass a custom descriptor upon connection (protocol option).

Edge

Start Edge with the --devtools-server-port option, for example:

MicrosoftEdge.exe --devtools-server-port 9222 about:blank

Please find more information here.

Firefox (Nightly)

Start Firefox with the --remote-debugging-port option, for example:

firefox --remote-debugging-port 9222

Bear in mind that this is an experimental feature of Firefox.

Bundled client

This module comes with a bundled client application that can be used to interactively control a remote instance.

Target management

The bundled client exposes subcommands to interact with the HTTP frontend (e.g., List, New, etc.), run with --help to display the list of available options.

Here are some examples:

$ chrome-remote-interface new 'http://example.com'
{
    "description": "",
    "devtoolsFrontendUrl": "/devtools/inspector.html?ws=localhost:9222/devtools/page/b049bb56-de7d-424c-a331-6ae44cf7ae01",
    "id": "b049bb56-de7d-424c-a331-6ae44cf7ae01",
    "thumbnailUrl": "/thumb/b049bb56-de7d-424c-a331-6ae44cf7ae01",
    "title": "",
    "type": "page",
    "url": "http://example.com/",
    "webSocketDebuggerUrl": "ws://localhost:9222/devtools/page/b049bb56-de7d-424c-a331-6ae44cf7ae01"
}
$ chrome-remote-interface close 'b049bb56-de7d-424c-a331-6ae44cf7ae01'

Inspection

Using the inspect subcommand it is possible to perform command execution and event binding in a REPL fashion that provides completion.

Here is a sample session:

$ chrome-remote-interface inspect
>>> Runtime.evaluate({expression: 'window.location.toString()'})
{ result: { type: 'string', value: 'about:blank' } }
>>> Page.enable()
{}
>>> Page.loadEventFired(console.log)
[Function]
>>> Page.navigate({url: 'https://github.com'})
{ frameId: 'E1657E22F06E6E0BE13DFA8130C20298',
  loaderId: '439236ADE39978F98C20E8939A32D3A5' }
>>> { timestamp: 7454.721299 } // from Page.loadEventFired
>>> Runtime.evaluate({expression: 'window.location.toString()'})
{ result: { type: 'string', value: 'https://github.com/' } }

Additionally there are some custom commands available:

>>> .help
[...]
.reset    Remove all the registered event handlers
.target   Display the current target

Embedded documentation

In both the REPL and the regular API every object of the protocol is decorated with the meta information found within the descriptor. In addition The category field is added, which determines if the member is a command, an event or a type.

For example to learn how to call Page.navigate:

>>> Page.navigate
{ [Function]
  category: 'command',
  parameters: { url: { type: 'string', description: 'URL to navigate the page to.' } },
  returns:
   [ { name: 'frameId',
       '$ref': 'FrameId',
       hidden: true,
       description: 'Frame id that will be navigated.' } ],
  description: 'Navigates current page to the given URL.',
  handlers: [ 'browser', 'renderer' ] }

To learn about the parameters returned by the Network.requestWillBeSent event:

>>> Network.requestWillBeSent
{ [Function]
  category: 'event',
  description: 'Fired when page is about to send HTTP request.',
  parameters:
   { requestId: { '$ref': 'RequestId', description: 'Request identifier.' },
     frameId:
      { '$ref': 'Page.FrameId',
        description: 'Frame identifier.',
        hidden: true },
     loaderId: { '$ref': 'LoaderId', description: 'Loader identifier.' },
     documentURL:
      { type: 'string',
        description: 'URL of the document this request is loaded for.' },
     request: { '$ref': 'Request', description: 'Request data.' },
     timestamp: { '$ref': 'Timestamp', description: 'Timestamp.' },
     wallTime:
      { '$ref': 'Timestamp',
        hidden: true,
        description: 'UTC Timestamp.' },
     initiator: { '$ref': 'Initiator', description: 'Request initiator.' },
     redirectResponse:
      { optional: true,
        '$ref': 'Response',
        description: 'Redirect response data.' },
     type:
      { '$ref': 'Page.ResourceType',
        optional: true,
        hidden: true,
        description: 'Type of this resource.' } } }

To inspect the Network.Request (note that unlike commands and events, types are named in upper camel case) type:

>>> Network.Request
{ category: 'type',
  id: 'Request',
  type: 'object',
  description: 'HTTP request data.',
  properties:
   { url: { type: 'string', description: 'Request URL.' },
     method: { type: 'string', description: 'HTTP request method.' },
     headers: { '$ref': 'Headers', description: 'HTTP request headers.' },
     postData:
      { type: 'string',
        optional: true,
        description: 'HTTP POST request data.' },
     mixedContentType:
      { optional: true,
        type: 'string',
        enum: [Object],
        description: 'The mixed content status of the request, as defined in http://www.w3.org/TR/mixed-content/' },
     initialPriority:
      { '$ref': 'ResourcePriority',
        description: 'Priority of the resource request at the time request is sent.' } } }

Chrome Debugging Protocol versions

By default chrome-remote-interface asks the remote instance to provide its own protocol.

This behavior can be changed by setting the local option to true upon connection, in which case the local version of the protocol descriptor is used. This file is manually updated from time to time using scripts/update-protocol.sh and pushed to this repository.

To further override the above behavior there are basically two options:

Browser usage

This module is able to run within a web context, with obvious limitations though, namely external HTTP requests (List, New, etc.) cannot be performed directly, for this reason the user must provide a global criRequest in order to use them:

function criRequest(options, callback) {}

options is the same object used by the Node.js http module and callback is a function taking two arguments: err (JavaScript Error object or null) and data (string result).

Using webpack

It just works, simply require this module:

const CDP = require('chrome-remote-interface');

Using vanilla JavaScript

To generate a JavaScript file that can be used with a <script> element:

  1. run npm install from the root directory;

  2. manually run webpack with:

     TARGET=var npm run webpack

  3. use as:

    <script>
  function criRequest(options, callback) { /*...*/ }
</script>
<script src="chrome-remote-interface.js"></script>

TypeScript Support

TypeScript definitions are kindly provided by Khairul Azhar Kasmiran and Seth Westphal, and can be installed from DefinitelyTyped:

npm install --save-dev @types/chrome-remote-interface

Note that the TypeScript definitions are automatically generated from the npm package devtools-protocol@0.0.927104. For other versions of devtools-protocol:

  1. Install patch-package using the instructions given.
  2. Copy the contents of the corresponding https://github.com/ChromeDevTools/devtools-protocol/tree/master/types folder (according to commit) into node_modules/devtools-protocol/types.
  3. Run npx patch-package devtools-protocol so that the changes persist across an npm install.

API

The API consists of three parts:

CDP([options], [callback])

Connects to a remote instance using the Chrome Debugging Protocol.

options is an object with the following optional properties:

These options are also valid properties of all the instances of the CDP class. In addition to that, the webSocketUrl field contains the currently used WebSocket URL.

callback is a listener automatically added to the connect event of the returned EventEmitter. When callback is omitted a Promise object is returned which becomes fulfilled if the connect event is triggered and rejected if the error event is triggered.

The EventEmitter supports the following events:

Event: 'connect'

function (client) {}

Emitted when the connection to the WebSocket is established.

client is an instance of the CDP class.

Event: 'error'

function (err) {}

Emitted when http://host:port/json cannot be reached or if it is not possible to connect to the WebSocket.

err is an instance of Error.

CDP.Protocol([options], [callback])

Fetch the Chrome Debugging Protocol descriptor.

options is an object with the following optional properties:

callback is executed when the protocol is fetched, it gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.Protocol((err, protocol) => {
    if (!err) {
        console.log(JSON.stringify(protocol, null, 4));
    }
});

CDP.List([options], [callback])

Request the list of the available open targets/tabs of the remote instance.

options is an object with the following optional properties:

callback is executed when the list is correctly received, it gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.List((err, targets) => {
    if (!err) {
        console.log(targets);
    }
});

CDP.New([options], [callback])

Create a new target/tab in the remote instance.

options is an object with the following optional properties:

callback is executed when the target is created, it gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.New((err, target) => {
    if (!err) {
        console.log(target);
    }
});

CDP.Activate([options], [callback])

Activate an open target/tab of the remote instance.

options is an object with the following properties:

callback is executed when the response to the activation request is received. It gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.Activate({id: 'CC46FBFA-3BDA-493B-B2E4-2BE6EB0D97EC'}, (err) => {
    if (!err) {
        console.log('target is activated');
    }
});

CDP.Close([options], [callback])

Close an open target/tab of the remote instance.

options is an object with the following properties:

callback is executed when the response to the close request is received. It gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.Close({id: 'CC46FBFA-3BDA-493B-B2E4-2BE6EB0D97EC'}, (err) => {
    if (!err) {
        console.log('target is closing');
    }
});

Note that the callback is fired when the target is queued for removal, but the actual removal will occur asynchronously.

CDP.Version([options], [callback])

Request version information from the remote instance.

options is an object with the following optional properties:

callback is executed when the version information is correctly received, it gets the following arguments:

When callback is omitted a Promise object is returned.

For example:

const CDP = require('chrome-remote-interface');
CDP.Version((err, info) => {
    if (!err) {
        console.log(info);
    }
});

Class: CDP

Event: 'event'

function (message) {}

Emitted when the remote instance sends any notification through the WebSocket.

message is the object received, it has the following properties:

Refer to the Chrome Debugging Protocol specification for more information.

For example:

client.on('event', (message) => {
    if (message.method === 'Network.requestWillBeSent') {
        console.log(message.params);
    }
});

Event: '<domain>.<method>'

function (params, sessionId) {}

Emitted when the remote instance sends a notification for <domain>.<method> through the WebSocket.

params is an object containing the payload.

sessionId is an optional string representing the session identifier.

This is just a utility event which allows to easily listen for specific notifications (see 'event'), for example:

client.on('Network.requestWillBeSent', console.log);

Additionally, the equivalent <domain>.on('<method>', ...) syntax is available, for example:

client.Network.on('requestWillBeSent', console.log);

Event: '<domain>.<method>.<sessionId>'

function (params, sessionId) {}

Equivalent to the following but only for those events belonging to the given session:

client.on('<domain>.<event>', callback);

Event: 'ready'

function () {}

Emitted every time that there are no more pending commands waiting for a response from the remote instance. The interaction is asynchronous so the only way to serialize a sequence of commands is to use the callback provided by the send method. This event acts as a barrier and it is useful to avoid the callback hell in certain simple situations.

Users are encouraged to extensively check the response of each method and should prefer the promises API when dealing with complex asynchronous program flows.

For example to load a URL only after having enabled the notifications of both Network and Page domains:

client.Network.enable();
client.Page.enable();
client.once('ready', () => {
    client.Page.navigate({url: 'https://github.com'});
});

In this particular case, not enforcing this kind of serialization may cause that the remote instance does not properly deliver the desired notifications the client.

Event: 'disconnect'

function () {}

Emitted when the instance closes the WebSocket connection.

This may happen for example when the user opens DevTools or when the tab is closed.

client.send(method, [params], [sessionId], [callback])

Issue a command to the remote instance.

method is a string describing the command.

params is an object containing the payload.

sessionId is a string representing the session identifier.

callback is executed when the remote instance sends a response to this command, it gets the following arguments:

When callback is omitted a Promise object is returned instead, with the fulfilled/rejected states implemented according to the error parameter. The Error object returned contains two additional parameters: request and response which contain the raw massages, useful for debugging purposes. In case of low-level WebSocket errors, the error parameter contains the originating Error object and no response is returned.

Note that the field id mentioned in the Chrome Debugging Protocol specification is managed internally and it is not exposed to the user.

For example:

client.send('Page.navigate', {url: 'https://github.com'}, console.log);

client.<domain>.<method>([params], [sessionId], [callback])

Just a shorthand for:

client.send('<domain>.<method>', params, sessionId, callback);

For example:

client.Page.navigate({url: 'https://github.com'}, console.log);

client.<domain>.<event>([sessionId], [callback])

Just a shorthand for:

client.on('<domain>.<event>[.<sessionId>]', callback);

When callback is omitted the event is registered only once and a Promise object is returned. Notice though that in this case the optional sessionId usually passed to callback is not returned.

When callback is provided, it returns a function that can be used to unsubscribe callback from the event, it can be useful when anonymous functions are used as callbacks.

For example:

const unsubscribe = client.Network.requestWillBeSent((params, sessionId) => {
    console.log(params.request.url);
});
unsubscribe();

client.close([callback])

Close the connection to the remote instance.

callback is executed when the WebSocket is successfully closed.

When callback is omitted a Promise object is returned.

client['<domain>.<name>']

Just a shorthand for:

client.<domain>.<name>

Where <name> can be a command, an event, or a type.

FAQ

Invoking Domain.methodOrEvent I obtain Domain.methodOrEvent is not a function

This means that you are trying to use a method or an event that are not present in the protocol descriptor that you are using.

If the protocol is fetched from Chrome directly, then it means that this version of Chrome does not support that feature. The solution is to update it.

If you are using a local or custom version of the protocol, then it means that the version is obsolete. The solution is to provide an up-to-date one, or if you are using the protocol embedded in chrome-remote-interface, make sure to be running the latest version of this module. In case the embedded protocol is obsolete, please file an issue.

See here for more information.

Invoking Domain.method I obtain Domain.method wasn't found

This means that you are providing a custom or local protocol descriptor (CDP({protocol: customProtocol})) which declares Domain.method while the Chrome version that you are using does not support it.

To inspect the currently available protocol descriptor use:

$ chrome-remote-interface inspect

See here for more information.

Why my program stalls or behave unexpectedly if I run Chrome in a Docker container?

This happens because the size of /dev/shm is set to 64MB by default in Docker and may not be enough for Chrome to navigate certain web pages.

You can change this value by running your container with, say, --shm-size=256m.

Using Runtime.evaluate with awaitPromise: true I sometimes obtain Error: Promise was collected

This is thrown by Runtime.evaluate when the browser-side promise gets collected by the Chrome's garbage collector, this happens when the whole JavaScript execution environment is invalidated, e.g., a when page is navigated or reloaded while a promise is still waiting to be resolved.

Here is an example:

$ chrome-remote-interface inspect
>>> Runtime.evaluate({expression: `new Promise(() => {})`, awaitPromise: true})
>>> Page.reload() // then wait several seconds
{ result: {} }
{ error: { code: -32000, message: 'Promise was collected' } }

To fix this, just make sure there are no pending promises before closing, reloading, etc. a page.

How does this compare to Puppeteer?

Puppeteer is an additional high-level API built upon the Chrome Debugging Protocol which, among the other things, may start and use a bundled version of Chromium instead of the one installed on your system. Use it if its API meets your needs as it would probably be easier to work with.

chrome-remote-interface instead is just a general purpose 1:1 Node.js binding for the Chrome Debugging Protocol. Use it if you need all the power of the raw protocol, e.g., to implement your own high-level API.

See #240 for a more thorough discussion.

Contributors

Resources