Claude Code: Interactive Google Vertex AI Setup Wizard
Claude CodeClaude Code v2.1.98 introduces an interactive setup wizard for Google Vertex AI, accessible from the login screen by selecting a 3rd-party platform option. The guided flow walks developers through GCP authentication, project and region selection, credential verification, and model pinning — eliminating the manual configuration steps that previously blocked enterprise teams from adopting Claude Code on Vertex AI.
Key Takeaways
- An interactive Vertex AI setup wizard is now built into the Claude Code login screen, replacing the previous documentation-only onboarding for GCP users.
- The guided flow covers all configuration steps — GCP authentication, project selection, region, credential verification, and model pinning — in a single in-tool experience.
- Model pinning via the wizard gives enterprise admins a governance lever to lock a specific Claude version across all Vertex AI deployments.
- Credential scrubbing from tool call results closes an edge case where environment variables echoed by shell commands could leak into model context.
- The --model flag regression is fixed, restoring correct model selection in headless and scripted Claude Code runs.
- MCP server initialization reliability improves for setups with multiple concurrent servers starting at session launch.
Sources & Mentions
5 external resources covering this update
New Features
Interactive Google Vertex AI Setup Wizard
Claude Code v2.1.98 ships an interactive setup wizard for Google Vertex AI, accessible directly from the login screen. When a user selects the 3rd-party platform option from the provider list, Claude Code now launches a step-by-step guided flow rather than redirecting to documentation. The wizard walks developers through GCP authentication, project selection, region configuration, credential verification, and optional model pinning — a significant reduction in friction for enterprise teams adopting Claude Code in GCP-governed environments.
Previously, connecting Claude Code to Vertex AI required manually setting environment variables, understanding GCP IAM roles, and correctly configuring the API endpoint — a process that was opaque enough to block adoption among developers unfamiliar with Google Cloud. The wizard addresses this by embedding the configuration steps directly into the tool's onboarding flow.
For teams deploying Claude Code at scale across Vertex AI, the wizard also supports model pinning, enabling admins to lock a specific Claude model version for all users — an important governance feature in regulated or cost-controlled environments.
Security Fixes
This release also addresses a credential exposure edge case: environment variables from tool call results are now scrubbed before being included in prompt context. This prevents secrets inadvertently echoed by shell commands (such as a printenv call or a script that outputs credentials) from being passed back to the model, reducing the risk of accidental credential leakage in agentic sessions.
Bug Fixes
Several smaller improvements ship alongside the Vertex AI wizard:
- The
--modelflag now correctly overrides the model for non-interactive (headless) runs, fixing a regression where model selection was ignored in scripted contexts. - An edge case in MCP server initialization has been resolved, improving reliability when multiple MCP servers start concurrently.
- Miscellaneous stability fixes for long-running sessions.