Hands-on course teaching deployment, management, and observability of AI agents on Kubernetes with KAgent, KMCP/MCP, system prompts, and integrations like Slack and AWS
The AI landscape is evolving quickly: agents are no longer limited to answering questions — they can take actions, connect to tools, and automate workflows. KAgent is a platform for running and managing these AI agents inside Kubernetes clusters. This course shows you how to design, deploy, and operate AI agents on Kubernetes using KAgent, KMCP/MCP, and related tooling.You will learn practical, production-focused topics:
How to install and operate KAgent in Kubernetes.
How agents communicate using the A2A (agent-to-agent) protocol.
How to build and run MCP servers and integrate them with agents.
How to create system prompts, define agents declaratively, and debug agent behavior using observability tools like Jaeger.
How to connect external tools and services (for example, Slack, AWS CloudWatch, Billing) to your agents.
Recommended prerequisites: basic Kubernetes knowledge (kubectl), familiarity with YAML manifests, and experience with Python or other scripting languages for building local integrations.
We begin with an overview of KAgent and a hands-on lab that walks you through the KAgent portal and agent lifecycle.
This section gives you a bird’s-eye view of the KAgent dashboard and how to interact with agents and tools through the UI. Next, we explain A2A — the secure agent-to-agent communication protocol that KAgent uses for agent interactions and orchestration.We then cover KAgent architecture and installation: core components, controllers, sidecar agents, and recommended deployment patterns. After installation you will validate the deployment and inspect running pods.
Example: after a successful installation you should see the KAgent-related pods running in the kagent namespace:
Once the foundation is ready, the course introduces KMCP and MCP servers. You will install a KMCP control plane and deploy one or more MCP servers (for example, an AWS-hosted MCP), enabling agents to query dynamic data sources or external APIs through the Model Context Protocol (MCP).
You will also build a custom MCP server. Example use case: agents fetch real-time cryptocurrency prices via an MCP endpoint so the agent can make data-driven decisions based on live market data.Example lab output when starting local components (CLI header):
Welcome to the KodeKloud Hands-On labKodeKloudAll rights reservedcontrolplane ~ via 🐍 v3.10.12
The course covers system prompts and prompt engineering for agents: how to craft system-level prompts, structure instructions for predictable behavior, and manage prompt templates used by multiple agents. You will also learn to declare agents using YAML manifests for reproducible deployments.Observability and debugging are core topics. We demonstrate tracing and telemetry with Jaeger to analyze end-to-end agent workflows, locate failures, and optimize performance. Example Jaeger Helm values for an all-in-one development deployment:
KMCP control plane, MCP server deployment, custom MCP
Build and connect MCP servers to agents
Agent Development
YAML manifests, system prompts, integrations
Create declarative and BYO agents
Observability & Debugging
Jaeger tracing, logs, metrics
Trace agent workflows and debug issues
Integrations & Labs
Slack bot, AWS CloudWatch, LangGraph example
Connect agents to real tools and services
Key components used in labs:
Component
Purpose
Notes / Examples
SLACK APP
External integration for notifications and commands
Create at api.slack.com/apps; needs SLACK_BOT_TOKEN, SLACK_APP_TOKEN
SLACK BOT (Python Bolt)
Local development bridge to KAgent
Runs python main.py, listens for /mykagent, formats Block Kit
KAGENT AGENT
Deployed in Kubernetes cluster
Declarative YAML manifests manage agent lifecycle
MCP Server
Provides contextual data to agents
Can be AWS-hosted or self-hosted; serve model context to LLMs
KMCP
Control plane for MCP servers
Coordinates MCP discovery and access
Security note: never commit tokens or secrets (for example SLACK_BOT_TOKEN, SLACK_APP_TOKEN) into source control. Use Kubernetes Secrets, environment variables at runtime, or a secrets manager.
Final hands-on lab: you will combine KAgent, MCP servers, system prompts, and observability to build a complete, deployable AI agent that integrates with external services like Slack and AWS CloudWatch. The course concludes with short quizzes after each section to reinforce key concepts.Links and references:
By the end of this course you will be able to design, deploy, and operate AI agents on Kubernetes with KAgent — managing communication, observability, and integrations in production-grade environments.
