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.
The following snippet loads https://github.com and dumps every request made:
const CDP = require('chrome-remote-interface');
CDP((client) => {
// extract domains
const {Network, Page} = client;
// setup handlers
Network.requestWillBeSent((params) => {
console.log(params.request.url);
});
Page.loadEventFired(() => {
client.close();
});
// enable events then start!
Promise.all([
Network.enable(),
Page.enable()
]).then(() => {
return Page.navigate({url: 'https://github.com'});
}).catch((err) => {
console.error(err);
client.close();
});
}).on('error', (err) => {
// cannot connect to the remote endpoint
console.error(err);
});Find more examples in the wiki, in particular notice how the above can be
rewritten using the async/await primitives.
You may also want to take a look at the FAQ.
npm install chrome-remote-interface
Install globally (-g) to just use the bundled client.
This module should work with every application implementing the Chrome Debugging Protocol. In particular, it has been tested against the following implementations:
| Implementation | Protocol version | Protocol | List | New | Activate | Close | Version |
|---|---|---|---|---|---|---|---|
| Google Chrome | tip-of-tree | yes | yes | yes | yes | yes | yes |
| Microsoft Edge | partial | yes | yes | no | no | no | yes |
| Node.js (v6.3.0+) | node | yes | no | no | no | no | yes |
| Safari (iOS) | partial | no | yes | no | no | no | no |
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.
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).
Start Chrome with the --remote-debugging-port option, for example:
google-chrome --remote-debugging-port=9222
Since version 59, additionally use the --headless option, for example:
google-chrome --headless --remote-debugging-port=9222
Plug the device and enable the port forwarding, for example:
adb forward tcp:9222 localabstract:chrome_devtools_remote
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>
Install and run the Edge Diagnostics Adapter.
Start Node.js with the --inspect option, for example:
node --inspect=9222 script.js
Install and run the iOS WebKit Debug Proxy.
This module comes with a bundled client application that can be used to interactively control a remote instance.
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'Using the inspect subcommand it is possible to
perform command execution
and event binding in a REPL fashion. But unlike
the regular API the callbacks are overridden to conveniently display the result
of the commands and the message of the events. Also, the event binding is
simplified here, executing a shorthand method (e.g., Page.loadEventFired())
toggles the event registration.
Remember that the REPL interface provides completion.
Here is a sample session:
$ chrome-remote-interface inspect
>>> Runtime.evaluate({expression: 'window.location.toString()'})
{ result:
{ result:
{ type: 'string',
value: 'https://www.google.it/_/chrome/newtab?espv=2&ie=UTF-8' },
wasThrown: false } }
>>> Page.enable()
{ result: {} }
>>> Page.loadEventFired() // registered
{ 'Page.loadEventFired': true }
>>> Page.loadEventFired() // unregistered
{ 'Page.loadEventFired': false }
>>> Page.loadEventFired() // registered
{ 'Page.loadEventFired': true }
>>> Page.navigate({url: 'https://github.com'})
{ result: { frameId: '28677.1' } }
{ 'Page.loadEventFired': { timestamp: 21385.383076 } }
>>> Runtime.evaluate({expression: 'window.location.toString()'})
{ result:
{ result: { type: 'string', value: 'https://github.com/' },
wasThrown: false } }To reduce the amount of data displayed by the event listeners it is possible to provide a filter function. In this example only the resource URL is shown:
$ chrome-remote-interface inspect
>>> Network.enable()
{ result: {} }
>>> Network.requestWillBeSent(params => params.request.url)
{ 'Network.requestWillBeSent': 'params => params.request.url' }
>>> Page.navigate({url: 'https://www.wikipedia.org'})
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/' }
{ result: { frameId: '5530.1' } }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/img/Wikipedia_wordmark.png' }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/img/Wikipedia-logo-v2.png' }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/js/index-3b68787aa6.js' }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/js/gt-ie9-c84bf66d33.js' }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/img/sprite-bookshelf_icons.png?16ed124e8ca7c5ce9d463e8f99b2064427366360' }
{ 'Network.requestWillBeSent': 'https://www.wikipedia.org/portal/wikipedia.org/assets/img/sprite-project-logos.png?9afc01c5efe0a8fb6512c776955e2ad3eb48fbca' }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-remote-interface uses the local version of the protocol descriptor by
default. This file is manually updated from time to time using
scripts/update-protocol.sh and pushed to this repository.
This behavior can be changed by setting the remote option to true
upon connection, in which case the remote instance is
asked to provide its own protocol descriptor.
Chrome < 60.0.3097.0 is not able to do that, so in that case the protocol descriptor is fetched from the source repository.
To override the above behavior there are basically three options:
-
pass a custom protocol descriptor upon connection (
protocoloption); -
use the raw version of the commands and events interface to use bleeding-edge features that do not appear in the local version of the protocol descriptor;
-
update the local copy with
scripts/update-protocol.sh(not present when fetched withnpm install).
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');To use a non-minified version manually run webpack with:
DEBUG=true npm run webpack
To generate a JavaScript file that can be used with a <script> element:
-
run
npm installfrom the root directory; -
manually run webpack with:
TARGET=var npm run webpack TARGET=var DEBUG=true npm run webpack -
use as:
<script> function criRequest(options, callback) { /*...*/ } </script> <script src="chrome-remote-interface.js"></script>
The API consists of three parts:
-
DevTools methods (for those implementations that support them, e.g., List, New, etc.);
-
connection establishment;
-
the actual protocol interaction.
Connects to a remote instance using the Chrome Debugging Protocol.
options is an object with the following optional properties:
-
host: HTTP frontend host. Defaults tolocalhost; -
port: HTTP frontend port. Defaults to9222; -
secure: HTTPS/WSS frontend. Defaults tofalse; -
target: determines which target this client should attach to. The behavior changes according to the type:- a
functionthat takes the array returned by theListmethod and returns a target or its numeric index relative to the array; - a target
objectlike those returned by theNewandListmethods; - a
stringrepresenting the raw WebSocket URL, in this casehostandportare not used to fetch the target list, yet they are used to complete the URL if relative; - a
stringrepresenting the target id.
Defaults to a function which returns the first available target according to the implementation (note that at most one connection can be established to the same target);
- a
-
protocol: Chrome Debugging Protocol descriptor object. Defaults to use the protocol chosen according to theremoteoption; -
remote: a boolean indicating whether the protocol must be fetched remotely or if the local version must be used. It has no effect if theprotocoloption is set. Defaults tofalse.
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:
function (client) {}Emitted when the connection to the WebSocket is established.
client is an instance of the CDP class.
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.
Fetch the Chrome Debugging Protocol descriptor.
options is an object with the following optional properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse;remote: a boolean indicating whether the protocol must be fetched remotely or if the local version must be returned. If it is not possible to fulfill the request then the local version is used. Defaults tofalse.
callback is executed when the protocol is fetched, it gets the following
arguments:
err: aErrorobject indicating the success status;protocol: an object with the following properties:remote: a boolean indicating whether the returned descriptor is the remote version or not (due to user choice or error);descriptor: the Chrome Debugging Protocol descriptor.
When callback is omitted a Promise object is returned.
For example:
const CDP = require('chrome-remote-interface');
CDP.Protocol(function (err, protocol) {
if (!err) {
console.log(JSON.stringify(protocol.descriptor, null, 4));
}
});Request the list of the available open targets/tabs of the remote instance.
options is an object with the following optional properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse.
callback is executed when the list is correctly received, it gets the
following arguments:
err: aErrorobject indicating the success status;targets: the array returned byhttp://host:port/json/listcontaining the target list.
When callback is omitted a Promise object is returned.
For example:
const CDP = require('chrome-remote-interface');
CDP.List(function (err, targets) {
if (!err) {
console.log(targets);
}
});Create a new target/tab in the remote instance.
options is an object with the following optional properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse;url: URL to load in the new target/tab. Defaults toabout:blank.
callback is executed when the target is created, it gets the following
arguments:
err: aErrorobject indicating the success status;target: the object returned byhttp://host:port/json/newcontaining the target.
When callback is omitted a Promise object is returned.
For example:
const CDP = require('chrome-remote-interface');
CDP.New(function (err, target) {
if (!err) {
console.log(target);
}
});Activate an open target/tab of the remote instance.
options is an object with the following properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse;id: Target id. Required, no default.
callback is executed when the response to the activation request is
received. It gets the following arguments:
err: aErrorobject indicating the success status;
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'}, function (err) {
if (!err) {
console.log('target is activated');
}
});Close an open target/tab of the remote instance.
options is an object with the following properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse;id: Target id. Required, no default.
callback is executed when the response to the close request is received. It
gets the following arguments:
err: aErrorobject indicating the success status;
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'}, function (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.
Request version information from the remote instance.
options is an object with the following optional properties:
host: HTTP frontend host. Defaults tolocalhost;port: HTTP frontend port. Defaults to9222;secure: HTTPS/WSS frontend. Defaults tofalse.
callback is executed when the version information is correctly received, it
gets the following arguments:
err: aErrorobject indicating the success status;info: a JSON object returned byhttp://host:port/json/versioncontaining the version information.
When callback is omitted a Promise object is returned.
For example:
const CDP = require('chrome-remote-interface');
CDP.Version(function (err, info) {
if (!err) {
console.log(info);
}
});function (message) {}Emitted when the remote instance sends any notification through the WebSocket.
message is the object received, it has the following properties:
method: a string describing the notification (e.g.,'Network.requestWillBeSent');params: an object containing the payload.
Refer to the Chrome Debugging Protocol specification for more information.
For example:
client.on('event', function (message) {
if (message.method === 'Network.requestWillBeSent') {
console.log(message.params);
}
});function (params) {}Emitted when the remote instance sends a notification for <domain>.<method>
through the WebSocket.
params is an object containing the payload.
This is just a utility event which allows to easily listen for specific
notifications (see 'event'), for example:
client.on('Network.requestWillBeSent', console.log);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', function () {
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.
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.
Issue a command to the remote instance.
method is a string describing the command.
params is an object containing the payload.
callback is executed when the remote instance sends a response to this
command, it gets the following arguments:
error: a boolean value indicating the success status, as reported by the remote instance;response: an object containing either the response (resultfield, iferror === false) or the indication of the error (errorfield, iferror === true).
When callback is omitted a Promise object is returned instead, with the
fulfilled/rejected states implemented according to the error parameter.
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);Just a shorthand for:
client.send('<domain>.<method>', params, callback);For example:
client.Page.navigate({'url': 'https://github.com'}, console.log);Just a shorthand for:
client.on('<domain>.<event>', callback);The only difference is that when callback is omitted the event is registered
only once and a Promise object is returned.
For example:
client.Network.requestWillBeSent(console.log);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.
This means that the protocol descriptor that you are using does not contain
Domain.method. If you are sure that your Chrome instance supports such method
you can call it directly:
client.send('Domain.method', ...);or ask Chrome the correct protocol descriptor:
CDP({remote: true});See here for more information.
This means that the protocol descriptor that you are using contains a method that is not supported by your Chrome instance. Most likely this is because you are trying to use a bleeding-edge feature, try to update to a newer Chrome version.
See here for more information.
To inspect the correct protocol descriptor use:
$ chrome-remote-interface inspect --remote
Bear in mind that --headless Chrome is relatively new and there are kinks (in
Chrome) that are being worked out. If you believe you have encountered a bug,
take a look at the open issues, especially external issues.
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.
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.
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.