Crossplane Agent

Deploy AI agents on Kubernetes — backed by Helm, powered by MCP, routed through AgentGateway.

Status: Single-agent setup is functional. Multi-agent is work in progress.

---

Prerequisites

Requirement	Details
Kubernetes cluster	kind / EKS / AKS / GKE
kubectl	latest stable
helm	>= 3
ArgoCD	installed in cluster
LLM API key	OpenAI / Gemini / Groq

---

Installation

0. Install ArgoCD (Required)

Create namespace:

kubectl create namespace argocd

Install ArgoCD:

kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

Access UI:

kubectl port-forward svc/argocd-server -n argocd 8080:443

Get admin password:

kubectl -n argocd get secret argocd-initial-admin-secret \
  -o jsonpath="{.data.password}" | base64 -d && echo

Login:

• URL: https://localhost:8080
• Username: admin
• Password: (from above command)

---

1. Install kagent CRDs

CRDs must be installed before the controller.

helm install kagent-crds oci://ghcr.io/kagent-dev/kagent/helm/kagent-crds \
  --namespace kagent \
  --create-namespace

Registers: Agent, ModelConfig, ToolServer, MCPIntegration

---

2. Install kagent

Export your API key:

export OPENAI_API_KEY="your-api-key"

Install:

helm install kagent oci://ghcr.io/kagent-dev/kagent/helm/kagent \
  --namespace kagent \
  --set providers.openAI.apiKey=$OPENAI_API_KEY

Optional — access the UI:

kubectl port-forward svc/kagent-ui -n kagent 8080:8080

Open http://localhost:8080

---

3. Install AgentGateway

Install Kubernetes Gateway API CRDs:

kubectl apply -f https://github.com/kubernetes-sigs/gateway-api/releases/latest/download/standard-install.yaml

Install AgentGateway CRDs and controller:

helm upgrade -i --create-namespace \
  --namespace agentgateway-system \
  --version v1.0.1 agentgateway-crds oci://cr.agentgateway.dev/charts/agentgateway-crds

helm upgrade -i -n agentgateway-system agentgateway oci://cr.agentgateway.dev/charts/agentgateway \
  --version v1.0.1

---

Deploy the Crossplane Agent

With ArgoCD in place, apply the App of Apps:

kubectl apply -f argocd/apps/apps-of-apps.yaml

This deploys:

• AgentGateway with the Crossplane MCP server as a backend
• The Crossplane MCP server itself
• kagent with an agent configured to consume it via RemoteMCPServer

---

Repository Structure

.
├── agentgateway/            # Gateway, HTTPRoute, CORS, and policy configs
├── argocd/                  # App of Apps — deploys everything via GitOps
├── mcp-server/              # Crossplane MCP server deployment and RBAC
├── single-agent/            # Single-agent setup (functional)
├── multi-agent/             # Multi-agent orchestration (not tested)
├── postgres/                # Database for agent registry state
└── skills/                  # Custom MCP tool definitions

Key files

argocd/
├── apps/apps-of-apps.yaml       # Entry point — apply this first
├── agentgateway-app.yaml
├── crossplane-app.yaml
├── crossplane-mcp-app.yaml
└── kagent-app.yaml

single-agent/
├── crossplane-agent.yaml        # kagent Agent definition
├── crossplane-mcp-server.yaml   # RemoteMCPServer config
├── agentgateway-crossplane-mcp.yaml
└── groq-modelconfig.yaml        # Example ModelConfig for Groq

agentgateway/
├── gateway.yaml
├── httproute.yaml
├── mcp-backend.yaml
├── crossplane-mcp-cors.yaml
└── agentgateway_policies.yaml

---

Multi-agent

Multi-agent manifests are present under multi-agent/ but are not tested.

---

Getting Started (TL;DR)

Once ArgoCD is installed:

kubectl apply -f argocd/apps/apps-of-apps.yaml

That’s it.

This will automatically:

• Install Crossplane
• Install AWS providers (S3, EC2)
• Deploy compositions (S3, VPC)
• Deploy MCP server
• Deploy kagent
• Configure AgentGateway routing

---

How It Works (High-Level)

User / Agent
     ↓
kagent (LLM reasoning)
     ↓
MCP Server (tools)
     ↓
Crossplane
     ↓
AWS Resources (S3, EC2, VPC)

If you want, you can use the predefined compositions, xrds, xrs

kubectl apply -f crossplane-resources

---

Example Use Case

Ask your agent:

"Create an S3 bucket for ML training data in ap-south-1"

Agent will:

1. Generate XR (S3Bucket)
2. Crossplane provisions AWS resource
3. Status returned via MCP

If you need predefined test scenarios

kubectl apply -f test_scenarios

---

Future Improvements

• Multi-cloud compositions (AWS + GCP + Azure)
• Crossplane Functions for policy enforcement
• Cost-aware provisioning
• Secure defaults (encryption, IAM)
• Production-ready multi-agent workflows

---

Contributing

PRs welcome — especially for:

• Multi-agent testing
• New MCP tools
• Crossplane compositions
• Observability integrations

---