Claude Code, Anthropic’s official terminal-based coding agent, just got a significant upgrade: a reference deployment for Google Cloud Platform.
The Gateway GCP example provides deployment assets for running Claude Code’s gateway layer on GCP infrastructure — Cloud Run, Secret Manager, and load-balanced behind Cloud Endpoints.
What the GCP Gateway Does
The Claude Gateway is the middleware layer that sits between Claude Code and your infrastructure. It handles:
- Authentication — verifies Claude Code sessions against your identity provider
- Secret management — injects credentials from GCP Secret Manager instead of local .env files
- Audit logging — records every agent action to Cloud Logging
- Network control — restricts which APIs and endpoints Claude Code can access
The GCP deployment rebrands this as “Agent Platform” — a clear signal that Anthropic sees Claude Code as an enterprise platform, not just a developer tool. The rebrand is covered in more detail in the companion piece on the Claude Gateway becoming a full Agent Platform.
Getting Started: What You Need
Before deploying, you’ll need:
- A GCP project with billing enabled
- The
gcloudCLI authenticated to your project - Claude Code installed locally (for testing against the gateway)
- Your Anthropic API key stored in GCP Secret Manager
The deployment is straightforward — all assets live in the Claude Code repository under the gateway examples directory:
cd claude-code/gateway-examples/gcp
gcloud run deploy claude-gateway \
--source . \
--region us-central1 \
--allow-unauthenticated # or configure IAP for authAfter deploy, point your local Claude Code at the gateway URL instead of hitting the Anthropic API directly. The gateway handles auth token exchange, so your developers never see raw API keys.
Why This Matters for Enterprises
Claude Code is already the most popular coding agent on the market, primarily because of Claude’s strong capabilities. But enterprise adoption requires infrastructure — security controls, audit trails, and integration with existing cloud environments.
The GCP Gateway addresses all three:
- Security: Secrets never touch the developer’s machine
- Audit: Every action is logged and searchable
- Control: Network policies restrict what the agent can access
Cost Considerations
Running the gateway on Cloud Run is pay-per-use — you only pay for actual request processing time. For a team of 10-20 developers using Claude Code daily, expect:
- Cloud Run: ~$5-20/mo for the gateway service itself
- Secret Manager: Negligible (pennies per secret version)
- Cloud Logging: Scales with audit volume; ~$0.50/GB ingested
- Cloud Endpoints: Free tier covers most team usage
The gateway doesn’t add API costs — those flow through to Anthropic as before. What it buys you is control: you can enforce model restrictions, rate limits, and cost caps at the gateway level before they reach your API budget.
How It Works in Detail
The deployment uses:
- Cloud Run for the gateway service (auto-scaling, pay-per-use)
- Secret Manager for credential storage
- Cloud Endpoints for API management and rate limiting
- Cloud Logging for audit trails
All assets are in the Claude Code repository under the gateway examples directory. Deploy with a single gcloud run deploy command.
This is the kind of infrastructure investment that separates “tools developers try” from “tools enterprises adopt.” The same enterprise-control story shows up elsewhere on the site — the Hermes security writeup on credential guards and provider isolation makes the same point about keeping secrets out of the agent’s hands, the Claude Code vs GitHub Copilot CLI comparison frames how different vendors approach the enterprise boundary, and the 2026 coding agent security checklist covers the full hardening checklist your team should run before deploying any agent in production. The GCP gateway is Anthropic’s answer: managed, logged, and network-restricted by default.