claudecode-vertex-proxy

🥈Silver

Proxy service enabling Claude Code to access Claude models via GCP Vertex AI. Ideal for operations teams needing secure, scalable AI model access. Integrates with existing workflows and tools.

24240Updated 3d ago

Intermediate30min to implementautomation

Saves ~210 min per use

Quick InstallView Source

git clone https://github.com/OrionStarAI/claudecode-vertex-proxy.git

Works with:

Claude

Overview

About This Skill

Proxy service enabling Claude Code to access Claude models via GCP Vertex AI. Ideal for operations teams needing secure, scalable AI model access. Integrates with existing workflows and tools.

How to Use

1. **Install the Proxy:** Run `npm install -g claudecode-vertex-proxy` or use the Docker image `ghcr.io/anthropics/claudecode-vertex-proxy:latest`. Configure the GCP service account with `roles/aiplatform.user` permissions. 2. **Configure Authentication:** Set environment variables `GOOGLE_APPLICATION_CREDENTIALS` to point to your service account key file. For production, use Workload Identity Federation instead of static keys. 3. **Route Requests:** Update your applications to send requests to `http://localhost:8080/v1/chat/completions` (default port) instead of direct Vertex AI endpoints. The proxy handles authentication and request formatting automatically. 4. **Monitor Performance:** Use the built-in metrics endpoint (`/metrics`) to track latency, error rates, and throughput. Integrate with Prometheus/Grafana for long-term monitoring. 5. **Scale Horizontally:** Deploy multiple proxy instances behind a load balancer for high availability. Configure health checks to `/healthz` endpoint every 30 seconds. **Pro Tips:** - For Claude Code integration, add the proxy URL to your `CLAUDE_CODE_CONFIG` file under `vertex_ai_endpoint`. - Enable request/response logging in debug mode during initial setup, then disable for production. - Use the `X-Request-ID` header to correlate logs across systems.

Setup & Installation

Quick Install

No install command available. Check the GitHub repository for manual installation instructions.

Alternative Install (Git Clone)

git clone https://github.com/OrionStarAI/claudecode-vertex-proxy

Requirements

Claude Code or compatible AI agent
Works with: Claude

Quick Start Guide

Install the Skill

Copy the install command above and run it in your terminal.

Open Your AI Agent

Launch Claude Code, Cursor, or your preferred AI coding agent.

Try It Out

Use the prompt template or examples below to test the skill.

Customize

Adapt the skill to your specific use case and workflow.

Usage Examples

Prompt Template

Set up a secure connection between [YOUR_ORGANIZATION]'s internal tools and GCP Vertex AI using the claudecode-vertex-proxy. Configure the proxy to route requests from [SPECIFIC_TOOL_OR_SYSTEM] through Vertex AI's [MODEL_NAME] endpoint. Test the connection by sending a sample prompt '[SAMPLE_PROMPT]' and verify the response latency is under [MAX_LATENCY_MS]ms. Document the setup steps and any required authentication credentials in [DOCUMENTATION_FILE].

Example Output

### Vertex AI Proxy Setup Report

**Configuration Summary:**
- Proxy Service: claudecode-vertex-proxy v1.2.3
- GCP Project: ops-prod-4217
- Vertex AI Endpoint: projects/ops-prod-4217/locations/us-central1/publishers/anthropic/models/claude-3-opus
- Internal Tool: Claude Code v1.15.0
- Authentication: Service Account `claude-proxy@ops-prod-4217.iam.gserviceaccount.com` with Vertex AI User role

**Connection Test Results:**
1. **Latency Benchmark:**
   - Average response time: 420ms (target: <500ms)
   - Peak latency: 680ms (during peak load at 14:22 UTC)
   - Status: ✅ Within SLA

2. **Sample Prompt Response:**
   **Prompt:** "Analyze the last 7 days of error logs from our Kubernetes cluster and identify the top 3 root causes of failures."
   **Response Quality:**
   - Detected 3 critical patterns:
     1. Memory pressure in namespace `monitoring` (OOMKilled containers)
     2. Certificate rotation failures in `istio-system` (expired certs)
     3. Database connection leaks in `postgres-operator` (unclosed connections)
   - Provided actionable remediation steps for each issue
   - Cited specific log file paths and timestamps

**Security Verification:**
- All requests routed through private VPC endpoint (10.128.0.0/28)
- IAM policies restrict access to service account only
- No sensitive data persisted in proxy logs

**Next Steps:**
- Schedule performance tuning for peak hours (14:00-16:00 UTC)
- Add monitoring dashboard for error rates and latency
- Document failover procedures for primary region (us-central1) to backup (europe-west1)