Step 8: Deploy to Cloud Run

One-Line Summary: Deploy your research agent to Google Cloud Run with a single ADK command — get a live HTTPS endpoint anyone can access.

Prerequisites: Complete agent with sessions from Step 7, a Google Cloud account

Why Cloud Run

Cloud Run is Google's serverless container platform. It is the natural deployment target for ADK agents because:

• Scales to zero — you pay nothing when nobody is using your agent
• Scales up automatically — handles traffic spikes without configuration
• HTTPS by default — every deployment gets a secure URL
• One-command deploy — ADK's CLI handles containerization, image push, and deployment

No Dockerfile to write. No container registry to manage. No Kubernetes to configure.

Prerequisites

Before deploying, you need:

1. A Google Cloud account — cloud.google.com (free tier available)
2. A Google Cloud project with billing enabled
3. The gcloud CLI installed — cloud.google.com/sdk/docs/install

Enable Required APIs

# Set your project
gcloud config set project YOUR_PROJECT_ID
 
# Enable the APIs that ADK needs
gcloud services enable \
  run.googleapis.com \
  artifactregistry.googleapis.com \
  secretmanager.googleapis.com \
  aiplatform.googleapis.com

Authenticate

# Log in to Google Cloud
gcloud auth login
 
# Set up application default credentials
gcloud auth application-default login

Deploy with One Command

ADK includes a built-in deployment command that handles everything:

adk deploy cloud_run \
  --project=YOUR_PROJECT_ID \
  --region=us-central1 \
  --service_name=research-agent \
  --with_ui \
  research_agent

That is it. One command. Here is what happens behind the scenes:

┌──────────────────────────────────────────────┐
│  adk deploy cloud_run                        │
│                                              │
│  1. Generates a Dockerfile from your code    │
│  2. Builds a container image                 │
│  3. Pushes to Artifact Registry              │
│  4. Deploys to Cloud Run                     │
│  5. Returns your live HTTPS URL              │
└──────────────────────────────────────────────┘

Command Flags

Access Your Agent

After deployment completes, you will see a URL like:

Service URL: https://research-agent-abc123-uc.a.run.app

If you used --with_ui, open that URL in your browser to get the same testing interface you used during development.

Call the API

Your agent is also available as an API. You can send requests programmatically:

# Send a research query to your deployed agent
curl -X POST https://research-agent-abc123-uc.a.run.app/run \
  -H "Content-Type: application/json" \
  -d '{
    "app_name": "research_agent",
    "user_id": "api_user",
    "session_id": "session_1",
    "message": "What are the top AI trends right now?"
  }'

Environment Variables

Your deployed agent needs the Gemini API key. Set it as a Cloud Run environment variable:

# Set environment variables on the deployed service
gcloud run services update research-agent \
  --region=us-central1 \
  --set-env-vars="GOOGLE_API_KEY=your-gemini-api-key"

Or pass extra gcloud flags during deployment using the -- separator:

adk deploy cloud_run \
  --project=YOUR_PROJECT_ID \
  --region=us-central1 \
  --service_name=research-agent \
  --with_ui \
  research_agent \
  -- \
  --set-env-vars="GOOGLE_API_KEY=your-gemini-api-key"

Update Your Deployment

When you make changes to your agent, redeploy with the same command:

# Same command — Cloud Run updates the existing service
adk deploy cloud_run \
  --project=YOUR_PROJECT_ID \
  --region=us-central1 \
  --service_name=research-agent \
  --with_ui \
  research_agent

Cloud Run performs zero-downtime deployments — it spins up the new version, routes traffic to it, then shuts down the old one.

Costs

Cloud Run pricing is based on usage:

For a development agent with light traffic, you will likely stay within the free tier. The main cost will be the Gemini API calls themselves.

Verify Your Deployment

Open your service URL in a browser. If you deployed with --with_ui, you should see the ADK web interface. Send a test message:

Search for the latest news about Google Cloud Run and summarize the top results.

If you see search results and a summary, your agent is live and working in production.

Key Takeaways

• adk deploy cloud_run handles containerization, image push, and deployment in one command
• No Dockerfile needed — ADK generates it for you
• --with_ui bundles the testing interface with your API
• Cloud Run scales to zero — you only pay for what you use
• Redeployment is the same command — zero-downtime updates

Reference: ADK Cloud Run Deployment · Cloud Run Pricing · Google Cloud Free Tier

← Sessions and Memory | Next: Step 9 - What's Next →