Claude inside Google Cloud
Vertex AI is Google Cloud's managed ML platform, and Claude is available through it. The reason to choose it mirrors Bedrock: teams already on Google Cloud can use Claude within their existing project, billing, IAM, and compliance — same models, GCP-native access.
Key concept · Same Claude, Google Cloud surface
As with Bedrock, what changes is the front door, not the model: authentication uses Google Cloud credentials, usage is billed to your GCP project, data stays in your Google Cloud boundary, and you deploy in Google Cloud regions. Everything you learned about building with Claude transfers — this course simply walks the same path through Vertex.
Enabling and calling Claude on Vertex
Vertex adds one setup step the direct API doesn't have: you enable the model in Model Garden first, then authenticate with the gcloud CLI.
Key concept · Model Garden, gcloud, and application-default credentials
The setup: (1) in the Google Cloud console, open Model Garden, search "Anthropic," and enable the Claude model you want. (2) Install the gcloud CLI and log in (gcloud init, gcloud auth login). (3) Set your project and create application-default credentials: gcloud config set project YOUR_PROJECT_ID and gcloud auth application-default login. After that, the Anthropic SDK (its Vertex mode, e.g. AnthropicVertex) automatically uses those credentials — you don't pass an API key — and you specify a project and region when constructing the client.
Everything you know — and a little more
Once connected, the whole toolkit carries over: requests, multi-turn, system prompts, temperature, streaming, structured data, evals, prompt engineering, tool use, RAG, features, MCP, and agents/workflows. The Google Cloud course is one of the most complete, and even goes a bit further than the base API course in a few places.
Key concept · Notable extras in the Vertex course
Alongside the standard arc, this course includes some deeper topics worth knowing: on the RAG side, reranking results and contextual retrieval (situating a chunk before embedding it) for better retrieval quality; on the tools side, the batch tool and tools for structured data; and on the apps side, parallelizing Claude Code, automated debugging, and computer use (with a "how computer use works" walkthrough). The platform-specific part is the setup above; the rest is Claude, the same as everywhere.
Chapter 24 quiz
Thirteen questions on Vertex setup, access, and what transfers. Answer first, then expand Show answer.
Part A · Multiple choice
Claude on Google Cloud is accessed through:
- A Vertex AI
- B Amazon Bedrock
- C The Anthropic Console only
- D Azure
Show answer
A — Vertex AI.
The first setup step unique to Vertex is:
- A Enable the Anthropic model in Model Garden
- B Request an Anthropic API key
- C Configure IAM roles in AWS
- D Nothing
Show answer
A.
Which command sets up application-default credentials?
- A gcloud auth application-default login
- B aws configure
- C npm install
- D /init
Show answer
A. After gcloud init / auth login and setting the project.
Once ADC is set up, how does the Anthropic SDK authenticate?
- A It automatically uses those Google Cloud credentials — no API key
- B You still pass an Anthropic API key
- C It uses IAM roles
- D It doesn't authenticate
Show answer
A. Via the Vertex mode (e.g. AnthropicVertex).
What do you specify when constructing the Vertex client?
- A A project and a region
- B An email
- C A temperature
- D Nothing
Show answer
A.
Which RAG topics does the Vertex course add beyond the base API course?
- A Reranking results and contextual retrieval
- B None
- C Prompt caching only
- D Streaming
Show answer
A.
Part B · True or false
Claude on Vertex is billed to your Google Cloud project.
Show answer
True. Within your GCP boundary.
The core building blocks (tool use, RAG, features, MCP, agents) transfer to Vertex.
Show answer
True. Same Claude; the setup is the platform-specific part.
You pass an Anthropic API key to call Claude on Vertex.
Show answer
False. You use Google Cloud application-default credentials.
The Vertex course also covers computer use and parallelizing Claude Code.
Show answer
True. Among its extra topics.
Contextual retrieval situates a chunk before embedding it.
Show answer
True. Improving retrieval quality.
Part C · Short answer
Give the four setup steps for Claude on Vertex.
Show answer
Enable the model in Model Garden; install/log in to the gcloud CLI; set your project; create application-default credentials (gcloud auth application-default login).
What is the one real difference between Bedrock and Vertex for a Claude developer?
Show answer
The platform's access/auth surface (AWS IAM vs. Google Cloud ADC + Model Garden) — the Claude capabilities are the same on both.