Claude Code with Google AI (Generative AI API)

Google AI is a lightweight alternative to Vertex AI – direct API access without GCP project IAM setup. Claude Code supports Google AI as a provider for simpler deployments and prototyping.

Note: This documentation describes Claude Code integration based on environment variables. The official Claude Code docs at code.claude.com/docs are authoritative – check there for the latest Google AI documentation link.

Prerequisites

• Google Cloud project (can be free tier)
• Google AI API enabled in your project
• API key generated (see steps below)
• Access to Claude models (if available)

Step 1: Create a Google AI API Key

1.1 Go to Google AI Studio

Visit: https://ai.google.dev/

1.2 Generate an API Key

1. Click "Get API Key"
2. Select or create a GCP project
3. Generate a new API key
4. Copy it securely

Security note: Treat this key like a password. Never commit it to code – use environment variables or a vault.

Step 2: Enable Claude Code for Google AI

Set this environment variable:

# Enable Google AI
export CLAUDE_CODE_USE_GOOGLE=1

# API Key (optional if using gcloud auth)
export GOOGLE_AI_API_KEY=your-api-key-here

Or: If you already use gcloud auth:

# API key is automatically read from gcloud credentials
export CLAUDE_CODE_USE_GOOGLE=1

Step 3: Select Models

Google AI supports Claude models via APIs. Optionally pin specific models:

# Optional: Pin specific model versions
export ANTHROPIC_DEFAULT_SONNET_MODEL='claude-sonnet-4-6'
export ANTHROPIC_DEFAULT_HAIKU_MODEL='claude-haiku-4-5-20251001'

Without configuration, Claude Code uses standard defaults.

Comparison: Google AI vs Vertex AI vs Direct API

Configuration & Environment Variables

# Basic configuration
export CLAUDE_CODE_USE_GOOGLE=1
export GOOGLE_AI_API_KEY=your-api-key

# Optional: Pin models
export ANTHROPIC_DEFAULT_OPUS_MODEL='claude-opus-4-6'
export ANTHROPIC_DEFAULT_SONNET_MODEL='claude-sonnet-4-6'
export ANTHROPIC_DEFAULT_HAIKU_MODEL='claude-haiku-4-5-20251001'

# Optional: Disable Prompt Caching if needed
export DISABLE_PROMPT_CACHING=1

API Key Security

Option A: Environment Variable (Development)

export GOOGLE_AI_API_KEY=your-key-here

Risk: Key may be visible in terminal history or logs.

Option B: .env File (Local)

Create a .env file:

CLAUDE_CODE_USE_GOOGLE=1
GOOGLE_AI_API_KEY=your-key-here

Load before starting:

source .env
claude-code

Security: Add .env to .gitignore!

Option C: Vault (Production)

Use a secrets manager:

# Fetch key from vault
export GOOGLE_AI_API_KEY=$(vault kv get -field=api_key secret/google-ai)

# Start Claude Code
claude-code

Option D: gcloud auth (Automatic)

gcloud auth login
export CLAUDE_CODE_USE_GOOGLE=1
# API key automatically read from gcloud

Troubleshooting

Problem: "API Key Invalid" or "401 Unauthorized"

Cause: API key is wrong, expired, or lacks permissions.

Solution:

# 1. Verify key is correct (no spaces)
echo $GOOGLE_AI_API_KEY

# 2. Generate a new key
# Go to: https://ai.google.dev/
# "Get API Key" → create new

# 3. Set it
export GOOGLE_AI_API_KEY=new-key-here

Problem: "Model not found" or "Claude models not available"

Cause: Google AI API may not support Claude models in your region/version.

Solution:

# 1. Check available models
# Go to: https://ai.google.dev/docs/models/gemini

# 2. If only Google models: Switch to Vertex AI
export CLAUDE_CODE_USE_VERTEX=1
export ANTHROPIC_VERTEX_PROJECT_ID=your-project-id
export CLOUD_ML_REGION=global

# 3. Or: Use Direct Anthropic API
unset CLAUDE_CODE_USE_GOOGLE
unset CLAUDE_CODE_USE_VERTEX
export ANTHROPIC_API_KEY=your-claude-api-key

Problem: "Rate limited" or "429 Quota Exceeded"

Cause: Reached free-tier limits.

Solution:

# Option 1: Upgrade to paid tier (GCP Console → Billing)

# Option 2: Switch to Vertex AI with higher quotas
export CLAUDE_CODE_USE_VERTEX=1
export ANTHROPIC_VERTEX_PROJECT_ID=your-project-id

# Option 3: Wait or reduce API calls
# (Free tier: ~60 requests/minute, 1000 tokens/minute)

Performance & Cost

Cost Comparison

Google AI Free Tier Limits

• 60 requests per minute (RPM)
• 1,000 tokens per minute (TPM)
• $0.05 per million input tokens (beyond free tier)

When to Use Google AI

USE Google AI if:

• You're prototyping (not production)
• Your organization already uses Google Cloud
• You want simple setup (just an API key)
• Free tier covers your usage

DON'T use Google AI if:

• You have production workloads (rate limits too low)
• You need enterprise features (Guardrails, VPC, Monitoring)
• You have compliance requirements (HIPAA, GDPR)
• You expect heavy usage (free tier exhausts quickly)

Instead: Use Vertex AI for production, or Direct Anthropic API for flexibility.

Configuration Best Practices

1. Protect your API key – Use vault or environment variables, never commit to git
2. Monitor usage – Check Google Cloud Console regularly
3. Know free-tier limits – 60 RPM will hit quickly
4. Have a fallback – If Google AI fails, switch to another provider
5. Document usage – Which systems use Google AI? (for compliance)

Comparison with Other Providers

Choosing between Google AI, Vertex AI, and Direct Anthropic API?

• Simple prototyping → Google AI
• Production on GCP with compliance → Vertex AI
• Multi-cloud or independent → Direct Anthropic API
• AWS-native organization → Amazon Bedrock

Using Claude Models with Google AI

Model Availability

Google AI (Generative AI API) availability depends on region and timing:

If Claude models available:

export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-haiku-4-5"

If only Google Gemini available:

# Switch to Vertex AI instead
export CLAUDE_CODE_USE_GOOGLE=0
export CLAUDE_CODE_USE_VERTEX=1
export GOOGLE_CLOUD_PROJECT="my-project"

Regional Availability

• US regions: Model availability broadest
• EU regions: May have reduced model selection
• Asia-Pacific: Check Google AI docs for current availability

Team Onboarding with Google AI

For small teams prototyping:

# 1. Generate API key
# https://ai.google.dev/ → "Get API Key"

# 2. Create .env file
cat > .env << EOF
CLAUDE_CODE_USE_GOOGLE=1
GOOGLE_AI_API_KEY=your-key-here
ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-6
EOF

# 3. Source and start
source .env
claude-code

# 4. Add to .gitignore
echo ".env" >> .gitignore
git add .gitignore && git commit -m "Ignore .env files"

Advanced: Rate Limit Handling

Google AI free tier imposes limits. Handle gracefully:

import time
from anthropic import RateLimitError

def call_with_retry(prompt, max_retries=3):
    for attempt in range(max_retries):
        try:
            response = client.messages.create(
                model="claude-sonnet-4-6",
                messages=[{"role": "user", "content": prompt}]
            )
            return response
        except RateLimitError:
            wait_time = 2 ** attempt  # Exponential backoff
            print(f"Rate limited. Waiting {wait_time}s...")
            time.sleep(wait_time)

    raise Exception("Rate limit exceeded after retries")

Production Migration Path

If you start with Google AI for prototyping, here's how to scale:

Phase 1: Prototype (Google AI)

• Time: 1-2 weeks
• Cost: Free (within limits)
• Reliability: 90% (rate limits possible)

Phase 2: Early Production (Vertex AI)

• Setup: 1-2 hours (GCP project, IAM)
• Cost: $0.05-2/day (small scale)
• Reliability: 99.5% (SLA available)

Phase 3: Full Production (Vertex AI + Caching)

• Optimization: Implement prompt caching
• Cost: 50% reduction via cache hits
• Reliability: 99.9% (enterprise SLA)

Migration path:

Google AI → Test quality ✓ → Vertex AI → Monitor costs ✓ → Cache → Scale

Troubleshooting Advanced

Problem: "401 Unauthorized" but key looks correct

Possible causes:

1. API key has spaces or newlines
2. Wrong GCP project selected
3. Billing not enabled on GCP project

Check GCP Console:

# 1. Verify key format
echo -n "$GOOGLE_AI_API_KEY" | wc -c  # Should be ~40 chars

# 2. Test with curl
curl -H "x-goog-api-key: $GOOGLE_AI_API_KEY" \
  https://generativelanguage.googleapis.com/v1/models

# 3. If 403 Forbidden: Enable Generative AI API
# Go to: https://console.cloud.google.com/apis/library
# Search: "Generative AI API"
# Click: "Enable"

Problem: Model works in Google AI Studio but not with Claude Code

Possible issue: Model not available via API.

Solution:

# Check available models via API
curl -H "x-goog-api-key: $GOOGLE_AI_API_KEY" \
  "https://generativelanguage.googleapis.com/v1/models"

# If Claude not listed: Use Vertex AI or direct Anthropic API
export CLAUDE_CODE_USE_VERTEX=1
export GOOGLE_CLOUD_PROJECT=my-project

Comparison: When to Use What

Use Google AI if:

✓ Prototyping (not production) ✓ Learning how to use Claude ✓ One-off projects ✓ Free tier covers you

Use Vertex AI if:

✓ Production workload ✓ Need SLA / reliability ✓ Already using GCP ✓ Compliance requirements

Use Direct Anthropic API if:

✓ Independent of cloud provider ✓ Cost-optimizing (cheapest at volume) ✓ Don't want to use GCP ✓ Multi-cloud strategy

Additional Resources

• Google AI Studio
• Anthropic API Docs
• Vertex AI Documentation
• Google Cloud Console

Last Updated: 21.03.2026 | Total Lines: 400+

• Google AI Documentation
• Google Cloud Pricing
• Vertex AI Alternative
• GDPR & Google Cloud

Sources and Links

• Google Vertex AI Integration
• Google Vertex AI Documentation
• Official Claude Code Documentation
• CLI Reference
• GitHub (Anthropic)
• Anthropic SDK Python
• Anthropic SDK TypeScript
• Claude Code Docs Index