Getting Started with Archestra: Deploy MCP Gateway on Kubernetes

📦 Deploy Archestra on Kubernetes: A Step-by-Step Tutorial

Let me start with a confession: I spent three hours debugging a YAML indentation error the first time I tried deploying Archestra. Three hours. So this tutorial is everything I wish I'd known from the start — no skipped steps, no "it's trivial" assumptions.

⚡ Prerequisites (What You Actually Need)

• Kubernetes cluster — any flavor works (Minikube, Kind, EKS, GKE, K3s). I tested on a 3-node K3s cluster with 4 vCPUs and 8GB RAM each.
• kubectl configured and pointing to your cluster
• Helm 3+ (Archestra ships with a Helm chart — this saves so much pain)
• Docker installed locally (for testing the image)

🔧 Step 1: Quick Docker Test (5 Minutes)

Before diving into Kubernetes, I recommend running the Docker image locally first. This confirms the image works and lets you explore the dashboard:

# Pull the image
docker pull archestra/platform:latest

# Run it (this takes about 20 seconds on a decent connection)
docker run -d \
  --name archestra \
  -p 8080:8080 \
  -v $(pwd)/archestra-data:/data \
  archestra/platform:latest

# Check the logs — you should see the gateway starting up
docker logs -f archestra

When you see Gateway listening on port 8080 in the logs, open http://localhost:8080. You'll see the Archestra dashboard. If you see a blank page — I hit this too — wait 10 more seconds. The dashboard needs a moment to initialize the MCP catalog.

☸️ Step 2: Deploy on Kubernetes with Helm (15 Minutes)

This is the production-grade setup. The Helm chart handles deployments, services, config maps, and ingress:

# Add the Archestra Helm repo
helm repo add archestra https://archestra-ai.github.io/helm-charts
helm repo update

# Create a namespace
kubectl create namespace archestra

# Install
helm install archestra archestra/archestra \
  --namespace archestra \
  --set image.tag=latest \
  --set ingress.enabled=true \
  --set ingress.host=archestra.example.com

🔌 Step 3: Connect Your First MCP Server (5 Minutes)

Once the gateway is running, connect an MCP server through the dashboard or API. Here's how I connected the GitHub MCP server:

# Register an MCP server via the Archestra API
curl -X POST https://archestra.example.com/api/mcp/servers \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "name": "github-mcp",
    "type": "mcp-server",
    "endpoint": "https://github-mcp.your-cluster.svc:8080",
    "auth_type": "token"
  }'

# List available servers to confirm
curl https://archestra.example.com/api/mcp/servers | jq '.'

Expected output: A JSON array with your newly registered MCP server. If you get an empty array, check that the MCP server pod is actually running (kubectl get pods -n archestra).

🧪 Performance: What I Measured

💭 Final Verdict

Archestra is the real deal for teams serious about MCP at scale. The security guardrails alone saved me from what would've been a production incident (one of my agents started making unauthorized API calls — Archestra's gateway blocked it). The 900+ curated MCP servers catalog is incredible for prototyping. For a non-hobbyist tool, the docs could be more beginner-friendly, but the Helm chart makes deployment smooth once you get past the initial config.