Demo Installing AWS Servers on KBs via KAgent UI

Hello — in this lesson we install and verify two AWS MCP (Managed Chat/Compute Provider) servers using KAgent: one via the KAgent UI and one via a manifest (YAML). We’ll:

Install a minimal KAgent deployment (with a provided 01-values-min).
Expose the KAgent UI.
Create an AWS Pricing MCP Server via the UI.
Create an AWS Well-Architected Security MCP Server via a manifest.
Verify the MCP servers and view their exposed tools.

Prerequisites: the CRD and model config for KAgent are already installed and the 01-values-min file has been provided.

The values file used below enables only the minimal components required to run KAgent and disables optional agents. This reduces resource usage while you work through the lab.

Minimal values file (`01-values-min`)

Use the following values to install KAgent with minimal components enabled:

cilium-debug-agent:
  enabled: false
cilium-manager-agent:
  enabled: false
cilium-policy-agent:
  enabled: false
helm-agent:
  enabled: false
istio-agent:
  enabled: false
k8s-agent:
  enabled: false
kgateway-agent:
  enabled: false
observability-agent:
  enabled: false
promql-agent:
  enabled: false

kmcp:
  enabled: true

kagent-tools:
  enabled: false

tools:
  grafana-mcp:
    enabled: false
  querydoc:
    enabled: false

ui:
  service:
    type: NodePort
    ports:
      port: 8080
      targetPort: 8080

Install KAgent using the provided values (installation command is assumed to be performed in the lab environment). After applying the chart/manifest, monitor the controller logs if you need to troubleshoot:

# Controller logs
kubectl -n kagent logs -l app.kubernetes.io/component=controller -f

Inspect KAgent pods and services

Confirm pods and services in the kagent namespace:

kubectl get pod -n kagent

Example output:

NAME                                              READY   STATUS    RESTARTS   AGE
kagent-controller-6886fc4f5c-4t7gd                0/1     ContainerCreating   0   16s
kagent-kmcp-controller-manager-76645f577f-fbqqs   1/1     Running             0   16s
kagent-ui-59d5bbd564-lssv4                        0/1     Running             0   16s

kubectl get svc -n kagent

Example output:

NAME                                                TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)                      AGE
kagent-controller                                   ClusterIP   172.20.160.95   <none>        8083/TCP                     30s
kagent-kmcp-controller-manager-metrics-service      ClusterIP   172.20.74.151   <none>        8443/TCP                     30s
kagent-ui                                           NodePort    172.20.42.19    <none>        8080:32165/TCP               30s

If you want to access the UI on a specific NodePort (for example 30080), patch the kagent-ui service:

kubectl patch svc kagent-ui -n kagent \
  -p '{"spec":{"type":"NodePort","ports":[{"name":"ui","port":8080,"targetPort":8080,"nodePort":30080}]}}'

After image pulls and initialization complete, pods should reach Running:

kubectl get pod -n kagent

Example final output:

NAME                                   READY   STATUS    RESTARTS   AGE
kagent-controller-6886fc4f5c-4t7gd     1/1     Running   0          63s
kagent-kmcp-controller-manager-76645f577f-fbqqs   1/1   Running   0   63s
kagent-ui-59d5bbd564-lssv4              1/1     Running   0          63s

AWS credentials used by MCP servers

MCP servers require AWS credentials (or other supported credential delivery methods). In this lab environment, credentials are stored at /root/.aws/credentials:

cat /root/.aws/credentials

Example output:

[default]
aws_access_key_id = AKIAWFKYLURA4GFSSD70
aws_secret_access_key = 5JG8kqNhMMSZMDbLjNJMYKqKS5rFxRPREqysVRPF

Region used in this lesson: us-east-1.

MCP servers support multiple credential delivery methods (plain keys, IAM roles, etc.). Follow security best practices for credential handling and avoid checking credentials into source control.

Install AWS Pricing MCP Server via the KAgent UI

Ensure the KAgent UI is accessible (port-forward or use the NodePort you set, e.g., 30080).
Open the KAgent UI in your browser (use the lab environment link or Node IP + NodePort).
In the UI:
- Click Create → New MCPServer.
- Click Add MCP Server.
- Enter the server name exactly: AWS Pricing MCP Server.
- Namespace: kagent (default).
- Choose the “Command” option (not URL).
- Command executor: select uvx (Python-based executor).
- Package name: awslabs.aws-pricing-mcp-server@latest.
- Provide environment variables copied from /root/.aws/credentials using these exact keys:
  - AWS_ACCESS_KEY_ID
  - AWS_SECRET_ACCESS_KEY
  - AWS_REGION (for example, us-east-1)

Example command used by the package:

uvx awslabs.aws-pricing-mcp-server@latest

After adding the server, KAgent creates a pod for the MCP Server. Monitor resources:

kubectl get pod -n kagent -w
kubectl get mcpserver -n kagent

You may first see the MCPServer with False readiness while images are pulling:

NAME                      READY   AGE
aws-pricing-mcp-server    False   51s

Wait until the MCPServer is True and the pod is Running:

kubectl get mcpserver -n kagent
kubectl get pod -n kagent

Example final output:

# kubectl get mcpserver -n kagent
# kubectl get pod -n kagent
# aws-pricing-mcp-server-6f59fd7dd8-5j7wh             1/1     Running   0          2m40s

If readiness stays False, inspect events, pod describe, and pod logs:

kubectl describe mcpserver aws-pricing-mcp-server -n kagent
kubectl describe pod <pod-name> -n kagent
kubectl logs <pod-name> -n kagent

Install AWS Well-Architected Security MCP Server via manifest

Create a file named mcp-server.yaml with the following MCPServer manifest. Populate AWS credentials in the env section before applying:

apiVersion: kagent.dev/v1alpha1
kind: MCPServer
metadata:
  name: awslabs-well-architected-security-mcp-server
  namespace: kagent
spec:
  deployment:
    args:
    - awslabs.well-architected-security-mcp-server
    cmd: uvx
    env:
      AWS_ACCESS_KEY_ID:
      AWS_REGION: us-east-1
      AWS_SECRET_ACCESS_KEY:
    image: ghcr.io/astral-sh/uv:debian
    port: 3000
  stdioTransport: {}
  transportType: stdio

Important: do not change the port (must remain 3000) or the transportType (stdio) — these are required by the package. Steps:

Edit mcp-server.yaml and set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY with values from /root/.aws/credentials.
Apply the manifest:

kubectl apply -f mcp-server.yaml

Example output:

mcpserver.kagent.dev/awslabs-well-architected-security-mcp-server created

Check status and pods:

kubectl get mcpserver -n kagent
kubectl get pod -n kagent

Example output:

# kubectl get mcpserver -n kagent
# NAME                                             READY   AGE
# aws-pricing-mcp-server                           True    6m31s
# kubectl get pod -n kagent
# NAME                                                            READY   STATUS    RESTARTS   AGE
# aws-pricing-mcp-server-6f59fd7dd8-5j7wh                         1/1     Running   0          6m44s
# awslabs-well-architected-security-mcp-server-8b68c79f9-v6x7v    1/1     Running   0          26s

Viewing tools exposed by an MCP Server in the UI

In the KAgent UI:

Click the MCPServer entry, then click View → Tools.
Each MCP Server lists how many tools it exposes. Example:
- AWS Well-Architected Security MCP Server — may show 6 tools.
- AWS Pricing MCP Server — may show 9 tools.
Click the number to inspect individual tools exposed by the MCP Server.

Notes and troubleshooting tips

Environment variables must use the exact key names required by the package: AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION.
uvx runs Python-based packages. Some packages may use npx (npm). Choose the executor that matches the package.
Image pulls and container initialization can take time. Use kubectl logs, kubectl describe, and kubectl get -w to monitor readiness.
For persistent issues, inspect controller logs:

kubectl -n kagent logs -l app.kubernetes.io/component=controller -f

Links and references

KAgent: Host Your AI Agents on Kubernetes (course): https://learn.kodekloud.com/user/courses/kagents-host-your-ai-agents-on-kubernetes
Kubernetes documentation: https://kubernetes.io/docs/
AWS CLI / Credentials best practices: https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-files.html

That’s it — you installed one MCP Server via the UI and one via a manifest, verified both, and viewed the tools each server exposes. See you in the next lab.

Introduction

KAgent Installation & Architecture Overview

KMCP Installation & Overview

System Prompts for AI Agent Building

Debugging AI Agents

Creating AI Agents - Bring It all together

Demo Installing AWS Servers on KBs via KAgent UI

Minimal values file (`01-values-min`)

Inspect KAgent pods and services

AWS credentials used by MCP servers

Install AWS Pricing MCP Server via the KAgent UI

Install AWS Well-Architected Security MCP Server via manifest

Viewing tools exposed by an MCP Server in the UI

Notes and troubleshooting tips

Links and references

Watch Video

Practice Lab

​Minimal values file (01-values-min)

​Inspect KAgent pods and services

​AWS credentials used by MCP servers

​Install AWS Pricing MCP Server via the KAgent UI

​Install AWS Well-Architected Security MCP Server via manifest

​Viewing tools exposed by an MCP Server in the UI

​Notes and troubleshooting tips

​Links and references

Watch Video

Practice Lab

Minimal values file (`01-values-min`)

Inspect KAgent pods and services

AWS credentials used by MCP servers

Install AWS Pricing MCP Server via the KAgent UI

Install AWS Well-Architected Security MCP Server via manifest

Viewing tools exposed by an MCP Server in the UI

Notes and troubleshooting tips

Links and references