Generate Temporary AWS Credentials

09 Jan, 2018
Xebia Background Header Wave

For our product Instruqt, we’re building challenges to learn AWS technology. Users of instruqt need to have temporary access to create, update and destroy their resources in AWS. We would like to give the users the experience of having their own AWS account. Access keys to use in the CLI or programmatic access, and the Management Console in the browser. This blog post describes how to prepare an AWS account for this usage, and how to generate and use temporary access keys, and a magic link to the management console.


  • AWS Account. You need to have an AWS Account, with a user and access keys with permissions to list, read, create, update and delete IAM groups, users, roles and policies.
  • Terraform. You need to have installed Terraform.
  • Python. Ensure you have a recent python version with pip. On a mac I recommend this setup.
  • AWS CLI. Install the AWS CLI with pip. (pip install awscli --upgrade --user)
  • Keybase. Install and create a Keybase account (

Roles and Federation

Besides common entities like users and groups, AWS provides “roles”. A role in AWS could for example be assigned to a server, which allows software running on the server to access AWS resources. A role can also be assumed by a user, giving him access to the resources. Also known as “federation”. A role contains two types of policies. One policy which describes the type of the service allowed to assume the role (an ec2 instance or an AWS account with users). The other policy describes the permission level to the specified resources.
For this use case we create a role which can be assumed by an AWS IAM User. Just a trust is not enough, the user also needs to have permissions to assume the role with AssumeRolePolicy. After the role is assumed, the AccessPolicy permissions are temporarliy given to the user by sending a set of keys.

|                  |
|       User       <-----------+
|                  |           |
+---------+--------+           |
          |                    |
          | AssumeRolePolicy   | TrustPolicy
          |                    |
+---------v--------+           |
|                  |           |
+       Role       +-----------+
|                  |
          | AccessPolicy 
|                  |
+     Resources    |
|                  |


To setup the user, role, policies etc, you could use CloudFormation, Terraform, the CLI or the Management Console. In this example I’ve used Terraform, but you could easily use this as documentation to do it manually, or using CloudFormation.
Create using the following source. You won’t have to change anything, but it’s recommended to read and search in the terraform documention what exactly happens.


variable "aws" {
  content = "Enter the aws profile to deploy."

variable "keybase" {
  content = "Enter the keybase profile to encrypt the secret_key (to decrypt: terraform output secret_key | base64 --decode | keybase pgp decrypt)"

variable "region" {
  default = "eu-west-1"

provider "aws" {
  profile    = "${}"
  region     = "${var.region}"

data "aws_caller_identity" "current" {}


resource "aws_iam_user" "instruqt" {
  name = "instruqt"

resource "aws_iam_access_key" "instruqt" {
  user = "${}"
  pgp_key = "keybase:${var.keybase}"

resource "aws_iam_user_policy" "instruqt_assume_role" {
  name = "test"
  user = "${}"

  policy = <<eof
  "Version": "2012-10-17",
  "Statement": [
      "Action": [
      "Effect": "Allow",
      "Resource": "*"

resource "aws_iam_role" "S3AccessRole" {
  name = "InstruqtS3Access"
  assume_role_policy = <<eof
  "Version": "2012-10-17",
  "Statement": [
      "Action": "sts:AssumeRole",
      "Principal": {
        "AWS": "arn:aws:iam::${data.aws_caller_identity.current.account_id}:root"
      "Effect": "Allow",
      "Sid": ""

resource "aws_iam_role_policy" "S3AccessPolicy" {
  name = "InstruqtS3AccessPolicy"
  role = "${}"
  policy = <<eof
  "Version": "2012-10-17",
  "Statement": [
      "Action": [
      "Effect": "Allow",
      "Resource": "*"


output "role_arn" {
  value = "${aws_iam_role.S3AccessRole.arn}"
output "access_key" {
  value = "${}"
output "secret_key" {
  value = "${aws_iam_access_key.instruqt.encrypted_secret}"

Deploy the stack using the following commands. Select one of the profiles configured in ~/.aws/credentials. Keep the profile name and your keybase username ready, because the apply command will ask you to enter these values.

terraform init
terraform apply

When Terraform has completed, the output contains access keys for the next steps. The secret key is encrypted with keybase. Decrypt the generated secret key using the next command:

terraform output secret_key | base64 --decode | keybase pgp decrypt

Run the python

First install dependencies with for example pip, using the next example. When you get errors of missing dependencies, install those as well.

pip install requests boto3

Create a file with the following source:

#!/usr/bin/env python
import os
import sys
import boto3
import json
import requests
import argparse
import ConfigParser
from botocore.exceptions import ClientError

def get_credentials_for_role(role_arn, session_name):
    sts = boto3.client('sts')
        response = sts.assume_role(RoleArn=role_arn, 
        return response['Credentials']
    except ClientError as e:
        sys.stderr.write('ERROR: %s\n' % e.response['Error']['Message'])

def write_credentials(profile, credentials):
    filename = os.path.expanduser('~/.aws/credentials')
    dirname = os.path.dirname(filename)

    if not os.path.exists(dirname):

    config = ConfigParser.ConfigParser()
    if not config.has_section(profile):
    config.set(profile, 'aws_access_key_id', credentials['AccessKeyId'])
    config.set(profile, 'aws_secret_access_key', credentials['SecretAccessKey'])
    config.set(profile, 'aws_session_token', credentials['SessionToken'])
    with open(filename, 'w') as fp:

def generate_console_link(credentials):
    session = json.dumps({'sessionId': credentials['AccessKeyId'],
                          'sessionKey': credentials['SecretAccessKey'],
                          'sessionToken': credentials['SessionToken']})

    r = requests.get("",
                     params={'Action': 'getSigninToken',
                             'SessionDuration': 43200,
                             'Session': session})
    signin_token = r.json()

    console = requests.Request('GET',
                              params={'Action': 'login',
                                      'Issuer': 'Instruqt',
                                      'Destination': '',
                                      'SigninToken': signin_token['SigninToken']})
    prepared_link = console.prepare()
    return prepared_link.url

if __name__ == '__main__':
    parser = argparse.ArgumentParser(content='generate keys')
    parser.add_argument("--output", "-o", required=False,
                        dest="output", help="output format", metavar="STRING",
                        default="json", choices=['link', 'json', 'write'])
    parser.add_argument("--role-arn", "-r", required=True,
                        dest="role_arn", help="to assume", metavar="STRING")
    parser.add_argument("--session-name", "-s", required=True,
                        dest="session_name", help="to use", metavar="STRING")

    options = parser.parse_args()
    credentials = get_credentials_for_role(options.role_arn, 

    if options.output == 'link':
        print generate_console_link(credentials)
    elif options.output == 'write':
        write_credentials(options.session_name, credentials)
    elif options.output == 'json':
        print(json.dumps({'AccessKeyId': credentials['AccessKeyId'],
                          'SecretAccessKey': credentials['SecretAccessKey'],
                          'SessionToken': credentials['SessionToken'],
                          'ConsoleMagicLink': generate_console_link(credentials)}))

The ./ script requires 3 parameters:

  • –session-name. The session name is an unique ID of the user who is going to use the temporary credentials. (Example:
  • –role-arn. The role arn is part of the output of the terraform script. (Example: arn:aws:iam::AWS_ACCOUNT_ID:role/InstruqtS3Access). You can copy this from the output of terraform.
  • –output. This could only contain: json | write | link, default output is json. uses Boto (AWS SDK for Python). It will use environment variables for access keys. Use the first example or copy the access_key and role_arn output from terraform, the decrypted secret_key and replace the example variables in the second example.
AWS_ACCESS_KEY_ID=$(terraform output access_key) \
AWS_SECRET_ACCESS_KEY=$(terraform output secret_key | base64 --decode | keybase pgp decrypt) \
python ./ --session-name identified@domain.ext \
                          --role-arn $(terraform output role_arn) \
                          --output json
AWS_SECRET_ACCESS_KEY=lk45hJSFkl35ADfsdDFtkl34fFADFhlktjrfaewr \
python ./ --session-name identified@domain.ext \
                          --role-arn arn:aws:iam::AWS_ACCOUNT_ID:role/InstruqtS3Access \
                          --output json

If you get an error with at the end of the stack trace the message: “The security token included in the request is invalid.” It’s probably because you didn’t replace the values of AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

Use the generated temporary keys

With a browser

Open a browser and copy the “console_access” magic link. You’re automatically logged on.
 Generate Temporary AWS Credentials

With the CLI

There are several ways to use the temporary credentials. To add the credentials once and easily use it in next commands, you create a new profile in ~/.aws/credentials. With --output write, the section is directly written into the credentials file and ready to be used.

aws s3 ls --profile tmpinstruqt

Programmatic access

And you can of course use these credentials for programmatic access. This is not the recommended way of adding secrets in your code, but just for this example hard coded.

import boto3
client = boto3.resource(
for bucket in client.buckets.all():

Final words

Now everything is setup and tested, just try to make it awesome. Give the temporary users more permissions to do other stuff than S3, generate and use the new keys in the CLI, scripts or using the Management Console.
To clean up, just remove the temporary credentials from your ~/.aws/credentials, and clear the terminal. Run terraform destroy to remove the AWS setup.
You can download the source from github:
Feel free to contact me if you have any questions. Your feedback is appreciated, and will be used to improve this blog post and future posts.
NB. Big thanks to my colleague Mark van Holsteijn. The python code used in the first version was copied from AWS Documentation. It wasn’t that clean. So we refactored it to the script now used in the blog post. It also gave us inspiration for new blog posts, so stay tuned for updates and related stuff.


Get in touch with us to learn more about the subject and related solutions

Explore related posts