Claude on Google Cloud (Anthropic managed): how to set up
2026-10-23 · 4 min read
Claude Models on Google Cloud (Anthropic managed) is the regular Claude API, run by Anthropic, that you reach through your Google Cloud project. Setup takes four steps: subscribe in Google Cloud Marketplace, enable two services, activate an Anthropic organization, then call Claude with your Google credentials instead of an API key.
The launch post came from @claudeai on October 22, which called it "the full set of Claude Platform features on Google Cloud," with Google Cloud IAM, billing and commitment retirement. Google's own announcement went up the same day.
How it differs from Claude on Vertex AI
Google Cloud customers could already use Claude through the Google-managed option, long known as Claude on Vertex AI. The catch was lag: new Claude Platform features showed up there only after Google built them into its own infrastructure.
With the Anthropic-managed option, Anthropic operates the service, so features like Claude Managed Agents, skills, code execution and the Files API work on day one. Google still handles identity, audit logs and the invoice. Spending counts toward any Google Cloud commitment you already have.
Here is how the two compare, according to Google's post and Anthropic's docs:
| Anthropic managed | Google managed | |
|---|---|---|
| Who processes your data | Anthropic, under its Commercial Terms | Google Cloud, under your existing agreement |
| New Claude Platform features | Day one | Once Google integrates them |
| Pricing | Standard Claude pricing | Standard Claude pricing |
| Regions | US and Global endpoints | US, EU and APAC data residency |
| Models | Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5 | Same |
My read: for a company that already runs on Google Cloud, this mostly removes the reason to keep a separate Anthropic account. You get a single bill and audit trail, and access lives in IAM next to everything else. The tradeoff is data processing. If a regulator or customer contract requires Google to be the processor, or you need EU data residency, stay on the Google-managed option.
How to set it up
Anthropic's setup guide goes through it in this order. You need permission to buy in Marketplace and to enable services on the project.
- In Google Cloud Marketplace, find the "Claude Models on Google Cloud (Anthropic managed)" listing, click Subscribe, then Manage on Console.
- Enable two services on each project that will call Claude:
gcloud services enable claude.googleapis.com \
anthropic-managed-claude.cloudpartnerservices.goog \
--project=my-project-id
- Click Activate on the Claude page, enter a contact email, follow the link it sends, fill in your organization details and accept Anthropic's terms.
- Click Open Claude Console on the same page and copy your workspace ID. It starts with
wrkspc_.
This creates a new Anthropic organization for the project. If your company already has a Claude Enterprise or API organization, the new one stays separate, and API keys, workspaces and stored files don't carry over.
Access is IAM roles, not API keys
You can't create API keys for this organization. Access comes from three project-level roles: roles/claude.viewer (read-only), roles/claude.user (send messages, run batches, manage files, skills and agents) and roles/claude.admin (everything, plus opening the Claude Console). For local development, grant yourself roles/claude.user, run gcloud auth application-default login, and set two environment variables:
export ANTHROPIC_GOOGLE_CLOUD_PROJECT="my-project-id"
export ANTHROPIC_GOOGLE_CLOUD_WORKSPACE_ID="wrkspc_..."
Your first call
Install the Python SDK with pip install -U "anthropic[google_cloud]" (the client is still in beta), then:
from anthropic import AnthropicGoogleCloud
client = AnthropicGoogleCloud()
message = client.messages.create(
model="claude-opus-5-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello, Claude"}],
)
print(message)
Model IDs match the regular Claude API. TypeScript uses @anthropic-ai/google-cloud-sdk. If you already have Claude code, swap the import and the client constructor, and leave the rest alone.
What it costs
Pricing is standard Claude pricing, billed on your Google Cloud invoice in Claude Consumption Units worth $0.01 each. Opus 5.5 fast mode works here too, at $8 per million input tokens and $40 per million output. If you have a negotiated discount, talk to your Anthropic or Google rep before you subscribe, because discounts can't be applied to past usage.
The docs list a few gaps: no API keys, no MCP tunnels, and the pre-built self-hosted sandbox workers in the SDKs don't support it yet. Revoking someone's Google access also won't stop a Managed Agents session or scheduled deployment that person already started. You have to interrupt or pause those yourself.
What this means for a business
Until now, a company on Google Cloud had to pick between its cloud contract and the newest Claude features. That choice mostly goes away here. A small team on Google Workspace and Google Cloud can run Claude agents with the same login controls and bill as everything else, so the procurement meeting gets much shorter.
What's left is deciding which process to hand Claude first. If you aren't sure, our free process audit maps where your team's hours go and shows which tasks are worth automating.