Google Cloud Run

Deploy to Google Cloud Run for serverless GCP deployment

Overview

Deploy the MCP Server with LangGraph to Google Cloud Run for fully managed, serverless hosting on Google Cloud Platform.

Serverless GCP: Auto-scaling from 0 to 100+ instances with pay-per-use pricing.

Quick Start

One-Command Deployment

cd cloudrun
./deploy.sh --setup

This automated script will:

1. Enable required GCP APIs
2. Create service account
3. Configure Secret Manager
4. Build Docker image
5. Deploy to Cloud Run
6. Output the service URL

Manual Deployment

## Build and deploy
gcloud run deploy mcp-server-langgraph \
  --source . \
  --region us-central1 \
  --allow-unauthenticated

Benefits

Serverless

Scale to zero when idle. Auto-scale to 100+ instances under load.

Pay-Per-Use

Only charged during request processing. No idle costs.

Automatic HTTPS

Free SSL certificates and automatic certificate renewal.

Secret Manager

Integrated with Google Cloud Secret Manager for secure configuration.

Fast Deployment

Deploy in 2-3 minutes with automatic rollback on failure.

Global

Deploy to regions worldwide for low latency.

Prerequisites

1. Google Cloud account with billing enabled
2. gcloud CLI installed: Install Guide
3. Docker (optional, for local testing)

Configuration

Secret Manager Setup

Store API keys securely:

## Run interactive setup
cd cloudrun
./setup-secrets.sh

Or manually:

## Create secrets
echo -n "your-jwt-secret" | gcloud secrets create jwt-secret-key --data-file=-
echo -n "sk-ant-..." | gcloud secrets create anthropic-api-key --data-file=-
echo -n "sk-..." | gcloud secrets create openai-api-key --data-file=-

Service Configuration

Edit cloudrun/service.yaml:

spec:
  template:
    metadata:
      annotations:
        autoscaling.knative.dev/minScale: '0'  # Scale to zero
        autoscaling.knative.dev/maxScale: '100'  # Max instances
    spec:
      containers:
      - resources:
          limits:
            cpu: '2'
            memory: 2Gi

Deployment

Initial Deployment

## Using script
cd cloudrun
./deploy.sh --setup

## Using gcloud
gcloud run deploy mcp-server-langgraph \
  --image gcr.io/PROJECT_ID/mcp-server-langgraph:latest \
  --region us-central1 \
  --platform managed \
  --allow-unauthenticated

Update Deployment

## Rebuild and redeploy
cd cloudrun
./deploy.sh

## Or
gcloud run deploy mcp-server-langgraph --source .

Rollback

## List revisions
gcloud run revisions list --service mcp-server-langgraph

## Rollback to previous revision
gcloud run services update-traffic mcp-server-langgraph \
  --to-revisions REVISION_NAME=100

Monitoring

View Logs

## Stream logs
gcloud run services logs tail mcp-server-langgraph --follow

## View recent logs
gcloud run services logs read mcp-server-langgraph --limit 100

Cloud Console

Access metrics in Cloud Console:

• Request count and latency
• CPU and memory utilization
• Instance count over time
• Error rates

Scaling

Configure Autoscaling

gcloud run services update mcp-server-langgraph \
  --min-instances 1 \
  --max-instances 100 \
  --concurrency 80

Strategies:

Scale to Zero (Cost-Optimized)

    --min-instances 0
    --cpu-throttling

• Pros: Lowest cost
• Cons: 1-2s cold start
• Use: Dev, low-traffic apps

Warm Pool (Low Latency)

    --min-instances 3
    --no-cpu-throttling

• Pros: No cold starts
• Cons: Higher baseline cost
• Use: Production, latency-sensitive

Hybrid (Balanced)

    --min-instances 1
    --max-instances 50

• Pros: Balance cost and latency
• Cons: Some cold starts during spikes
• Use: Most production workloads

Security

Network Security

## Restrict to internal traffic only
gcloud run services update mcp-server-langgraph \
  --ingress internal

## Or allow only from load balancer
gcloud run services update mcp-server-langgraph \
  --ingress internal-and-cloud-load-balancing

VPC Access

Connect to private resources:

## Create VPC connector
gcloud compute networks vpc-access connectors create mcp-connector \
  --region us-central1 \
  --network default \
  --range 10.8.0.0/28

## Use connector
gcloud run services update mcp-server-langgraph \
  --vpc-connector mcp-connector \
  --vpc-egress private-ranges-only

Authentication

## Require authentication
gcloud run services update mcp-server-langgraph \
  --no-allow-unauthenticated

## Grant access to specific users
gcloud run services add-iam-policy-binding mcp-server-langgraph \
  --member="user:[e-mail]" \
  --role="roles/run.invoker"

Cost Optimization

Estimated costs: $5-30/month for typical usage (10K requests/day)

Optimization tips:

1. Scale to zero when idle (--min-instances 0)
2. CPU throttling for cost savings
3. Right-size resources (start with 1 CPU, 1Gi memory)
4. Request bundling to reduce request count charges
5. Caching to reduce LLM API calls

Troubleshooting

503 Service Unavailable

Causes:

• Container startup timeout
• Health check failures
• Insufficient memory

Solutions:

# Increase timeout and memory
gcloud run services update mcp-server-langgraph \
  --timeout 600 \
  --memory 4Gi

Secret Access Denied

Cause: Service account lacks permissions

Solution:

gcloud secrets add-iam-policy-binding SECRET_NAME \
  --member="serviceAccount:SERVICE_ACCOUNT_EMAIL" \
  --role="roles/secretmanager.secretAccessor"

Cold Start Latency

Solutions:

• Use --min-instances 1 or higher
• Enable startup CPU boost (default in service.yaml)
• Optimize Docker image size

Complete Guide

This page provides comprehensive Cloud Run deployment instructions. For additional deployment options:

Production Checklist

Pre-deployment security and compliance checklist

Monitoring Setup

Configure observability for Cloud Run

Next Steps

Deploy Now

Follow quick start guide above

Configure Monitoring

Set up observability and alerts

Production Checklist

Pre-deployment security checklist

Compare Platforms

Choose the right deployment option

---

Ready to deploy? Run cd cloudrun && ./deploy.sh --setup to deploy to Google Cloud Run!