• If you’re running in Compute Engine or App Engine, authentication should “just work”.

  • If you’re developing locally, the easiest way to authenticate is using the Google Cloud SDK:

    $ gcloud beta auth application-default login

    Note that this command generates credentials for client libraries. To authenticate the CLI itself, use:

    $ gcloud auth login

    Previously, gcloud auth login was used for both use cases. If your gcloud installation does not support the new command, please update it:

    $ gcloud components update
  • If you’re running your application elsewhere, you should download a service account JSON keyfile and point to it using an environment variable:

    $ export GOOGLE_APPLICATION_CREDENTIALS="/path/to/keyfile.json"

Client-Provided Authentication#

Every package uses a Client as a base for interacting with an API. For example:

from import datastore
client = datastore.Client()

Passing no arguments at all will “just work” if you’ve followed the instructions in the Overview. The credentials are inferred from your local environment by using Google Application Default Credentials.

Credential Discovery Precedence#

When loading the Application Default Credentials, the library will check properties of your local environment in the following order:

  1. Application running in Google App Engine
  2. JSON or PKCS12/P12 keyfile pointed to by GOOGLE_APPLICATION_CREDENTIALS environment variable
  3. Credentials provided by the Google Cloud SDK (via gcloud auth login)
  4. Application running in Google Compute Engine

Explicit Credentials#

The Application Default Credentials discussed above can be useful if your code needs to run in many different environments or if you just don’t want authentication to be a focus in your code.

However, you may want to be explicit because

  • your code will only run in one place
  • you may have code which needs to be run as a specific service account every time (rather than with the locally inferred credentials)
  • you may want to use two separate accounts to simultaneously access data from different projects

In these situations, you can create an explicit Credentials object suited to your environment. After creation, you can pass it directly to a Client:

client = Client(credentials=credentials)

Google App Engine Environment#

To create credentials just for Google App Engine:

from oauth2client.contrib.appengine import AppAssertionCredentials
credentials = AppAssertionCredentials([])

Google Compute Engine Environment#

To create credentials just for Google Compute Engine:

from oauth2client.contrib.gce import AppAssertionCredentials
credentials = AppAssertionCredentials([])

Service Accounts#

A service account can be used with both a JSON keyfile and a PKCS12/P12 keyfile.

Directly creating credentials in oauth2client for a service account is a rather complex process, so as a convenience, the from_service_account_json() and from_service_account_p12() factories are provided to create a Client with service account credentials.

For example, with a JSON keyfile:

client = Client.from_service_account_json('/path/to/keyfile.json')


Unless you have a specific reason to use a PKCS12/P12 key for your service account, we recommend using a JSON key.

User Accounts (3-legged OAuth 2.0) with a refresh token#

The majority of cases are intended to authenticate machines or workers rather than actual user accounts. However, it’s also possible to call Google Cloud APIs with a user account via OAuth 2.0.


A production application should use a service account, but you may wish to use your own personal user account when first getting started with the google-cloud-python library.

The simplest way to use credentials from a user account is via Application Default Credentials using gcloud auth login (as mentioned above):

from oauth2client.client import GoogleCredentials
credentials = GoogleCredentials.get_application_default()

This will still follow the precedence described above, so be sure none of the other possible environments conflict with your user provided credentials.

Advanced users of oauth2client can also use custom flows to create credentials using client secrets or using a webserver flow. After creation, Credentials can be serialized with to_json() and stored in a file and then and deserialized with from_json().


Setting up a Service Account#

If your application is not running on Google Compute Engine, you need a Google Developers Service Account.

  1. Visit the Google Developers Console.

  2. Create a new project or click on an existing project.

  3. Navigate to APIs & auth > APIs and enable the APIs that your application requires.


You may need to enable billing in order to use these services.

  • BigQuery
    • BigQuery API
  • Datastore
    • Google Cloud Datastore API
  • Pub/Sub
    • Google Cloud Pub/Sub
  • Storage
    • Google Cloud Storage
    • Google Cloud Storage JSON API
  1. Navigate to APIs & auth > Credentials.

    You should see a screen like one of the following:

Find the “Add credentials” drop down and select “Service account” to be guided through downloading a new JSON keyfile.

If you want to re-use an existing service account, you can easily generate a new keyfile. Just select the account you wish to re-use, and click Generate new JSON key:

Using Google Compute Engine#

If your code is running on Google Compute Engine, using the inferred Google Application Default Credentials will be sufficient for retrieving credentials.

However, by default your credentials may not grant you access to the services you intend to use. Be sure when you set up the GCE instance, you add the correct scopes for the APIs you want to access:

  • All APIs

  • BigQuery

  • Datastore

  • Pub/Sub

  • Storage


Advanced Customization#

Though the google-cloud-python library defaults to using oauth2client to sign requests and httplib2 for sending requests, it is not a strict requirement.

The Client constructor accepts an optional http argument in place of a credentials object. If passed, all HTTP requests made by the client will use your custom HTTP object.

In order for this to be possible, the http object must do two things:

The entire signature from httplib2 need not be implemented, we only use it as

http.request(uri, method=method_name, body=body, headers=headers)

For an example of such an implementation, a google-cloud-python user created a custom HTTP class using the requests library.

As for handling authentication on your own, it may be easiest just to re-use bits from oauth2client. Unfortunately, these parts have a hard dependency on httplib2. We hope to enable using custom HTTP libraries with oauth2client at some point.