Claude Code Deployment

Overview

Claude Code supports deployment through multiple providers beyond the direct Claude API, including Amazon Bedrock and Google Vertex AI for enterprise cloud deployment.

Amazon Bedrock Integration

Overview

Claude Code integrates with Amazon Bedrock to enable deployment through AWS infrastructure using Claude models available in your AWS account.

Prerequisites

• Active AWS account with Bedrock access enabled
• Access to desired Claude models (e.g., Claude Sonnet 4.5)
• AWS CLI installed (optional)
• Appropriate IAM permissions

Setup Process

1. Model Access

Navigate to the Amazon Bedrock console, access Model access settings, and request Claude model availability in your region.

2. AWS Credentials Configuration

Multiple authentication methods are supported:

AWS CLI:

aws configure

Environment variables:

export AWS_ACCESS_KEY_ID=your-key
export AWS_SECRET_ACCESS_KEY=your-secret
export AWS_SESSION_TOKEN=your-token  # Optional

SSO profile:

aws sso login --profile=<name>
export AWS_PROFILE=your-profile

Bedrock API keys:

export AWS_BEARER_TOKEN_BEDROCK=your-token

3. Claude Code Configuration

Enable Bedrock integration:

export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION=us-east-1  # Or preferred region

Optional override for Haiku region:

export ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION=us-west-2

4. Model Selection

Default models include Claude Sonnet 4.5 and Claude Haiku 4.5.

Customize via:

export ANTHROPIC_MODEL='model-id'
export ANTHROPIC_SMALL_FAST_MODEL='haiku-model-id'

5. Token Configuration

Recommended settings:

export CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096
export MAX_THINKING_TOKENS=1024

IAM Permissions

Required actions:

• bedrock:InvokeModel
• bedrock:InvokeModelWithResponseStream
• bedrock:ListInferenceProfiles

Example IAM policy:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream",
        "bedrock:ListInferenceProfiles"
      ],
      "Resource": "*"
    }
  ]
}

Advanced Features

Automatic credential refresh supports corporate identity providers through awsAuthRefresh and awsCredentialExport configuration options.

Key Limitations

• Login/logout commands disabled (AWS credentials handle authentication)
• Uses Bedrock's Invoke API, not Converse API

Google Vertex AI Integration

Overview

Claude Code integrates with Google Vertex AI to enable deployment through Google Cloud Platform. The service supports both global and regional endpoints for model access.

Prerequisites

• Active GCP account with billing enabled
• A project with Vertex AI API access
• Google Cloud SDK (gcloud) installed
• Appropriate quota allocation in your chosen region

Setup Process

1. Enable Vertex AI API

Enable the Vertex AI API in your GCP project:

gcloud config set project YOUR-PROJECT-ID
gcloud services enable aiplatform.googleapis.com

2. Request Model Access

Navigate to Vertex AI Model Garden to search for and request access to Claude models like Claude Sonnet 4.5.

Approval time: Typically 24-48 hours

3. Configure GCP Credentials

Claude Code uses standard Google Cloud authentication and automatically detects the project ID from environment variables.

gcloud auth application-default login

4. Configure Claude Code

Set environment variables:

export CLAUDE_CODE_USE_VERTEX=1
export CLOUD_ML_REGION=global  # Or specify regional endpoints
export ANTHROPIC_VERTEX_PROJECT_ID=YOUR-PROJECT-ID

5. Model Configuration

Default models include Claude Sonnet 4.5 as the primary model and Claude Haiku 4.5 as the fast model.

Customize through environment variables:

export ANTHROPIC_MODEL='model-id'
export ANTHROPIC_SMALL_FAST_MODEL='haiku-model-id'

Key Features

Prompt Caching: Automatically supported via cache_control flags

1M Token Context: Available in beta for Sonnet 4 and 4.5

IAM Requirements: Assign roles/aiplatform.user role for necessary permissions:

gcloud projects add-iam-policy-binding YOUR-PROJECT-ID \
  --member="user:email@example.com" \
  --role="roles/aiplatform.user"

Troubleshooting

Quota limitations:

• Check quota in GCP Console
• Request increases if needed

Unsupported models in specific regions:

• Verify model availability in Model Garden
• Switch to supported regional endpoints

429 rate-limit errors:

• Implement retry logic
• Request quota increases
• Spread requests across regions

Comparison: Bedrock vs Vertex AI vs Claude API

Feature	Claude API	AWS Bedrock	Google Vertex AI
Setup Complexity	Simple	Moderate	Moderate
Authentication	API key	AWS credentials	GCP credentials
Regional Options	Global	AWS regions	GCP regions
Billing	Direct	AWS billing	GCP billing
Enterprise Features	Basic	Advanced	Advanced
Compliance	Standard	AWS compliance	GCP compliance

Best Practices for Enterprise Deployment

1. Use OIDC/Workload Identity for credential management
2. Implement quota monitoring to avoid service interruptions
3. Set up proper IAM roles with least privilege access
4. Configure region preferences based on data residency requirements
5. Enable logging and monitoring for audit trails
6. Use environment-specific configurations for dev/staging/prod
7. Implement cost controls with budget alerts
8. Test failover scenarios between regions
9. Document credential rotation procedures
10. Review security policies regularly

CI/CD Integration

Both Bedrock and Vertex AI support automated workflows:

GitHub Actions with Bedrock:

- name: Configure AWS Credentials
  uses: aws-actions/configure-aws-credentials@v1
  with:
    role-to-assume: arn:aws:iam::ACCOUNT:role/ROLE
    aws-region: us-east-1

- name: Run Claude Code
  run: |
    export CLAUDE_CODE_USE_BEDROCK=1
    claude -p "task" --output-format json

GitLab CI with Vertex AI:

script:
  - gcloud auth activate-service-account --key-file=$GCP_KEY_FILE
  - export CLAUDE_CODE_USE_VERTEX=1
  - export ANTHROPIC_VERTEX_PROJECT_ID=$PROJECT_ID
  - claude -p "task"