KCD Toronto 2026

Session Overview on MCP Servers, Agents, and Kubernetes Notes

The session focused on selecting the right operators, gateways, and tooling for running MCP servers and AI agents on Kubernetes. The presentation explored the following:

• Running MCP servers securely on Kubernetes
• Managing AI agents in production environments
• Choosing between Kubernetes-native operators and gateways
• Safety, lifecycle management, scalability, and integrations for agentic workloads

The central theme throughout the talk was the following:

Kubernetes is becoming the operating system for AI agents and MCP infrastructure.

Key Concepts Covered

1. MCP (Model Context Protocol) on Kubernetes

What is MCP?

MCP (Model Context Protocol) enables AI agents and LLMs to interact with tools, APIs, infrastructure, and external systems in a structured and standardized way.

Examples include:

• Accessing Kubernetes clusters
• Querying databases
• Calling internal APIs
• Performing DevOps operations
• Reading logs and metrics

The session emphasized that

MCP servers are becoming the bridge between AI agents and operational infrastructure.

2. Kubernetes MCP Server

The presentation highlighted:

• **containers/kubernetes-mcp-server**

Purpose

The Kubernetes MCP Server provides AI agents with secure and controlled access to Kubernetes clusters through MCP interfaces.

It is designed to be production-ready and safety-focused.

Important Features

• Multi-cluster support
• RBAC-aware access
• Safety controls and guardrails
• Action confirmations
• Controlled operational scope
• Evaluation and testing for safer agent behavior

Why It Matters

Without MCP abstractions, AI agents may execute unrestricted commands directly against Kubernetes clusters.

The Kubernetes MCP server introduces the following:

• Safer interaction models
• Limited operational boundaries
• Controlled actions
• Improved governance and auditing

3. kubernetes-mcp-server vs kubectl

One of the most important discussions compared

• **kubernetes-mcp-server**
• **kubectl**

Kubernetes MCP Server

Advantages

• Safer for production environments
• Built specifically for AI agent safety
• Restricts available actions
• Supports approval workflows
• Stronger RBAC integration
• Easier auditing and governance

Best Use Cases

• Production AI agents
• Enterprise automation
• Guardrailed operations
• Multi-tenant environments

kubectl

Advantages

• Extremely flexible
• Full Kubernetes functionality
• Integrates with many tools and CLIs

Risks

• Dangerous for autonomous agents
• Assumes trusted human operators
• Minimal safety layers
• Easy to perform destructive operations accidentally

Best Use Cases

• Human operators
• Advanced debugging
• Engineering and administrative workflows

4. MCP Lifecycle Operator

The talk introduced:

• **kubernetes-sigs/mcp-lifecycle-operator**

Purpose

The MCP Lifecycle Operator manages the lifecycle of MCP servers on Kubernetes.

It is emerging as a standard approach for:

• Deploying MCP servers
• Managing upgrades and updates
• Scaling MCP infrastructure
• Standardizing operational workflows

When to Use the MCP Lifecycle Operator

1. Containerized MCP Servers

Use it when:

• MCP servers run inside containers
• Kubernetes-native deployment patterns are required

2. External Integrations

Useful for:

• Integrating with Kubernetes ecosystem projects
• Connecting gateways, observability platforms, policies, and service meshes

3. Large-Scale Deployments

Best suited for:

• Enterprise-scale MCP environments
• Multi-team operations
• Consistent operational governance

Additional Ecosystem Projects

The session also referenced several emerging Kubernetes + AI ecosystem projects worth understanding.

5. Kuadrant / MCP Gateway

• **kuadrant/mcp-gateway**

Purpose

Acts as a gateway layer for MCP traffic and agent communications.

Think of it as:

An API gateway specifically designed for MCP-based workloads.

What It Provides

Security

• Authentication
• Authorization
• Rate limiting
• Policy enforcement

Routing

• Routes MCP requests to appropriate services and tools

Multi-Tenancy

• Separates workloads, agents, and teams

Observability

• Metrics
• Logging
• Tracing

Governance

• Centralized policy and communication control

Why It Matters

As organizations deploy larger numbers of AI agents:

• Traffic management becomes critical
• Security boundaries become mandatory
• Centralized governance becomes essential

MCP gateways may eventually become:

The “Ingress Controller” equivalent for AI agents.

6. AgentGateway

• **agentgateway**

Purpose

A unified gateway platform for AI agents and agentic workloads.

Focus Areas

• Agent-to-agent communication
• Tool routing
• Context management
• Security
• Agent orchestration

Why It’s Important

Traditional API gateways were designed primarily for:

• Human-facing applications
• REST APIs
• Standard web traffic

AgentGateway introduces patterns tailored for:

• Autonomous agents
• Long-running workflows
• Tool invocation
• Multi-agent coordination

7. Kagent

• **kagent**

Purpose

A Kubernetes-native framework/operator for running AI agents.

Think of it as:

Kubernetes primitives for agent workflows.

What Kagent Enables

Agent Deployment

• Run agents as Kubernetes-native resources

Declarative Management

• Manage agents through YAML and CRDs

Scaling

• Horizontally scale agents

Kubernetes Integration

Native integration with:

• Services
• Secrets
• RBAC
• Events
• Jobs
• Workflows

Observability

Track:

• Agent behavior
• Tool usage
• Execution history

Why Kagent Matters

Kagent pushes AI agents into becoming:

First-class Kubernetes workloads.

Just as:

• Deployments manage applications
• Jobs manage batch processing

Kagent aims to manage the following:

• Autonomous AI agents

8. Kubernetes SIGs Agent Sandbox

• **kubernetes-sigs/agent-sandbox**

Purpose

An experimental environment for Kubernetes-native AI agent projects.

The sandbox appears focused on:

• Safe experimentation
• Agent runtime patterns
• Governance models
• Standardization efforts

Why the Sandbox Matters

The Kubernetes + AI ecosystem is still in its early stages.

The sandbox enables:

• Rapid experimentation
• Community collaboration
• Standard definitions
• Testing emerging patterns

Bigger Industry Trend

The session strongly reflected a broader industry movement:

Kubernetes + AI Convergence

The industry is rapidly moving toward the following:

• AI agents operating infrastructure
• AI-native platform engineering
• Autonomous DevOps workflows
• Agent orchestration platforms

Key Takeaway

The Kubernetes ecosystem is rapidly evolving beyond container orchestration into a platform for managing intelligent, autonomous systems.

The combination of:

• MCP servers
• Agent gateways
• Lifecycle operators
• Kubernetes-native agent frameworks

is laying the foundation for:

AI-native infrastructure operations and autonomous platform engineering.