Agentic Kubernetes CLI

An intelligent Agentic Kubernetes CLI assistant that integrates your local MCP server with AWS Bedrock, enabling natural language control over kubectl commands, cluster management, and real-time Kubernetes insights.

---

🚀 Features

🗣️ Natural Language Kubernetes Operations

• Conversational kubectl interface — execute commands in plain English
• AI-powered command generation and validation
• Context-aware management of clusters and namespaces
• Real-time monitoring: pods, logs, resources, and events

⚙️ Advanced Capabilities

• ✅ Resource Management — List, describe, and manage Kubernetes resources
• 📊 Troubleshooting Assistant — Analyze logs and diagnose issues
• 🔄 Context Switching — Move between clusters easily
• ⚡ Interactive CLI — Friendly prompt-driven interface

---

🎥 Demo

Below is an example interaction between the User and the Agentic Kubernetes CLI demonstrating real-time troubleshooting and remediation.

---

🧑‍💻 User: list all pods in the cluster
🤖 Agent: There are no running pods, but one pod nginx is showing an issue (ImagePullBackOff).

🧑‍💻 User: analyze the issue with nginx pod
🤖 Agent: Issue identified — the container image tag nginx:lt is incorrect and not found in the registry.

🧑‍💻 User: fix the image tag issue and make the pod running
🤖 Agent: Updated the image tag to nginx:latest, redeployed the pod, and verified that it is now Running successfully.

---

Demo.mp4

---

📚 Table of Contents

• Prerequisites
• Installation
• Configuration
• Usage
• Examples
• Architecture
• MCP Tools

---

🔧 Prerequisites

Required Software

• Python 3.12+
• kubectl
• Access to a running Kubernetes cluster
• AWS account with Bedrock API access

AWS Bedrock Setup

1. Enable Claude models in the AWS Bedrock console
2. Configure AWS credentials with Bedrock access
3. Ensure IAM permissions for bedrock:InvokeModel

---

📦 Installation

Quick Setup

# Clone the repository
git clone https://github.com/Tarique-B-DevOps/Agentic-Kubernetes-CLI.git
cd Agentic-Kubernetes-CLI

# Install dependencies
pip install -r requirements.txt

# Verify kubectl installation
kubectl version --client

# Run the agent
python3 agent.py

---

⚙️ Configuration

Environment Variables

Set or export the following variables before running the agent:

# AWS Bedrock Configuration
export BEDROCK_MODEL_REGION=us-east-1
export BEDROCK_MODEL_ID=apac.anthropic.claude-3-5-sonnet-20241022-v2:0

# AWS Credentials (optional if configured with AWS CLI)
export AWS_ACCESS_KEY_ID=your_access_key
export AWS_SECRET_ACCESS_KEY=your_secret_key
export AWS_REGION=us-east-1

---

🎯 Usage

Start the Agent

python3 agent.py

Example Interactive Commands

⎈ kubectl> list all pods in production namespace
⎈ kubectl> what's wrong with my nginx deployment?
⎈ kubectl> show me the last 100 logs from api-server
⎈ kubectl> switch to staging cluster

---

💡 Examples

List Pods

⎈ kubectl> show all pods in the cluster

AI Response:

✅ Command executed successfully:
NAME                     READY   STATUS    RESTARTS   AGE
nginx-7854ff8877-2kxq9   1/1     Running   0          5d
api-server-abc123        1/1     Running   2          3d

Troubleshoot a Pod

⎈ kubectl> why is my app-server pod failing?

AI Response:

Analyzing pod app-server...
Issue: ImagePullBackOff
Root Cause: Container exits with code 1
Last Error: Connection refused to database:5432
Suggestion: Check database service and credentials

View Logs

⎈ kubectl> get the last 50 lines of logs from nginx pod and summarize

Manage Contexts

⎈ kubectl> list all available clusters
⎈ kubectl> switch to production-cluster
⎈ kubectl> what's my current context?

Scale Deployments

⎈ kubectl> scale nginx deployment to 5 replicas

Namespace Operations

⎈ kubectl> show all resources in kube-system namespace
⎈ kubectl> get all deployments across all namespaces

---

🏗️ Architecture

System Components

---

🧩 MCP Tools

run_kubectl_command

{
  "command": "kubectl get pods -n default -o json",
  "return_code": 0,
  "stdout": "...",
  "success": true
}

kubectl_context

{
  "action": "list",
  "success": true,
  "output": "..."
}