Open WebUI Kubernetes Integration

TL;DR

Chat with your Kubernetes cluster using natural language!

This project connects OpenWebUI to Kubernetes, letting you manage your cluster through conversational AI. Ask questions like "What pods are running?" or "Scale my deployment to 5 replicas" and get instant results.

---

Table of Contents

• TL;DR
  • Quick Start
• Configure Open-WebUI
  • Step 1: Access Open-WebUI
  • Step 2: Add the MCP Bridge Tool Server
  • Step 3: Add Connections to Models
  • Step 4: Verify Tool Integration
  • Step 5: Test the Integration
• Troubleshooting Configuration
• Architecture Overview
  • Architecture Diagram
  • Data Flow
  • Network Architecture
• Repository Structure
• Components
  • 1. Open WebUI
  • 2. Ollama
  • 3. MCP-Bridge
  • 4. Kubernetes Cluster
• Usage

---

Quick Start

One-command setup:

./install.sh

Componentes installed:

• Kind (Kubernetes in Docker)
• Open Web UI
• Ollama
• MCP bridge

After successful installation, you'll see: ✅ Docker Compose startup complete!

Services available at:

• Open WebUI: http://localhost:3000
• MCP Bridge: http://localhost:9000
• Ollama API: http://localhost:11434

Configure Open WebUI to use the MCP tools

Configure Open-WebUI

After running the installation script, you need to configure Open-WebUI to use the Kubernetes tools:

Step 1: Access Open-WebUI

1. Open your browser and go to http://localhost:3000
2. Create an admin account on first visit
3. Sign in with your new account

Step 2: Add the MCP Bridge Tool Server

1. Click on your profile icon (top right)
2. Go to Settings
3. Navigate to Admin Panel → Tools
4. Click "+ Add Tool Server"
5. Enter the following details:
   • Name: Kubernetes Tools
   • URL: http://mcpo:9000
   • Description: Kubernetes management via kubectl and helm
6. Click "Add"

---

---

Step 3: Add Connections to Models

Configure your AI model connections - use the local Ollama or add remote connections:

1. In Settings, go to Admin Panel → Connections
2. For Local Ollama (recommended):
   • The local Ollama server should be automatically detected
   • Verify connection to http://ollama:11434
3. For Remote Connections (optional):
   • Add OpenAI, Anthropic, or other model providers
   • Configure API keys and endpoints as needed

Step 4: Verify Tool Integration

1. Go back to the chat interface
2. Look for the tools icon (🔧) in the chat input area
3. You should see available tools like:
   • kubectl_get - List Kubernetes resources
   • kubectl_apply - Apply manifests
   • kubectl_describe - Describe resources
   • helm_install - Install Helm charts
   • And more...

Step 5: Test the Integration

Try asking these questions to verify everything works (copy and paste into the chat):

Copy & Paste these queries:

What pods are running in the kube-system namespace?

Show me all services in the default namespace

List all deployments across all namespaces

Create a new namespace called production

Install Jenkins using Helm in the jenkins namespace

Troubleshooting Configuration

If tools don't appear:

• Check that the MCP Bridge is running: curl http://localhost:9000/health
• Verify the tool server URL is exactly: http://mcpo:9000
• Restart Open-WebUI: docker-compose restart open-webui

If you get connection errors:

• Ensure all containers are running: docker-compose ps
• Check container logs: docker-compose logs mcpo
• Verify network connectivity between containers

Then open http://localhost:3000 and start chatting with your cluster!

Example queries:

• "What pods are running in kube-system?"
• "Show me all services in default namespace"
• "Create a new namespace called production"
• "Install Jenkins using Helm in the jenkins namespace"

---

Architecture Overview

This project integrates Open WebUI with Kubernetes management capabilities through a bridge architecture that connects multiple components to provide AI-powered Kubernetes operations.

Architecture Diagram

graph TB

    subgraph "External AI Services"
        Gemini[Google Gemini
External API]
    end
    
    subgraph "User Interface"
        User[User]
    end
    
    subgraph "UI Layer"
        OpenWebUI[Open WebUI
Port: 3000]
        Ollama[Ollama
LLM Server
Port: 11434]
    end
    
    subgraph "Bridge Layer"
        MCP-Bridge[MCP-Bridge
OpenAPI Bridge
Port: 9000]
    end
    
    subgraph "Target Infrastructure"
        K8s[Kubernetes Cluster
kubectl, helm, istioctl]
    end
    
    subgraph "Configuration"
        KubeConfig[kubeconfig
./kube/config]
    end
    


    %% User interactions
    User -->|Chat & Commands| OpenWebUI
    
    %% AI Layer connections
    OpenWebUI -->|LLM Requests| Ollama
    OpenWebUI -.->|External API Calls| Gemini
    OpenWebUI -->|Tool Calls
OpenAPI/REST| MCP-Bridge
    
    %% Bridge Layer to Kubernetes
    MCP-Bridge -->|kubectl commands| K8s
    MCP-Bridge -->|helm operations| K8s
    MCP-Bridge -->|istioctl commands| K8s
    
    %% Configuration
    KubeConfig -.->|Mounted Volume| MCP-Bridge
    
    %% Styling
    classDef userLayer fill:#e1f5fe
    classDef aiLayer fill:#f3e5f5
    classDef externalLayer fill:#fff8e1
    classDef bridgeLayer fill:#fff3e0
    classDef mcpLayer fill:#e8f5e8
    classDef k8sLayer fill:#fce4ec
    classDef configLayer fill:#f1f8e9
    
    class User userLayer
    class OpenWebUI,Ollama aiLayer
    class Gemini,OpenAI,Anthropic externalLayer
    class MCP-Bridge bridgeLayer
    class K8s k8sLayer
    class KubeConfig configLayer

---

Data Flow

1. User Query: User asks a Kubernetes-related question in Open WebUI
2. AI Processing: Ollama processes the query and determines if tools are needed
3. Tool Selection: Open WebUI identifies the appropriate kubectl_get tool
4. API Call: Open WebUI makes REST API call to MCPO bridge
5. Protocol Translation: MCPO translates REST call to MCP protocol
6. Response Chain: Results flow back through the same chain to the user

Network Architecture

graph LR
    subgraph "Docker Network: mcp-lab-network"
        subgraph "External Access"
            Host[Host Machine
localhost:3000
localhost:9000]
        end
        
        subgraph "Internal Services"
            OW[open-webui:8080]
            MC[mcpo:9000]
            K8S[Kubernetes:6443]
            OL[ollama:11434]
        end
    end
    
    Host -->|Port 3000| OW
    Host -->|Port 9000| MC
    OW --> OL
    OW --> MC
    MC --> K8S

---

📂 Repository Structure

openwebui-k8s-bridge/
├── mcp-bridge/          # Bridge service code
├── scripts/             # Setup and utility scripts
├── tests/               # Test scripts and examples
├── docs/                # Documentation
├── kube/               # Kubernetes configuration
├── docker-compose.yml  # Complete stack setup
└── README.md           # Main documentation

Components

1. Open WebUI

• Purpose: Web-based chat interface for AI interactions
• Port: 3000
• Role: Provides the user interface and orchestrates AI conversations with tool calling capabilities
• Key Features:
  • Chat interface for natural language Kubernetes queries
  • Tool server integration for external API calls
  • Model management and conversation history

2. Ollama

• Purpose: Local LLM server
• Port: 11434
• Role: Provides the AI language model (llama3.2:latest) for understanding user queries and generating responses
• Key Features:
  • Local model hosting
  • Function calling capabilities
  • Integration with Open WebUI

3. MCP Bridge

• Purpose: Protocol bridge between OpenAPI and MCP
• Port: 9000
• Role: Translates REST API calls from Open WebUI into MCP protocol calls
• Key Features:
  • OpenAPI specification generation for Open WebUI
  • REST to MCP protocol translation
  • Tool parameter validation and formatting

4 Kubernetes Cluster

• Purpose: Target infrastructure for management operations
• Role: The actual Kubernetes cluster being managed
• Access: Through kubeconfig mounted as volume

Usage

Once configured, you can ask natural language questions about your Kubernetes cluster:

• "What pods are running in the kube-system namespace?"
• "Show me all services in the default namespace"
• "List all deployments across all namespaces"
• "Get the logs from the nginx pod"
• install argocd using helm from repo https://argoproj.github.io/argo-helm

---

The AI will automatically use the appropriate Kubernetes tools to execute commands and provide formatted responses.

Open in Google Cloud Shell

You can directly open this repository in Google Cloud Shell to start exploring the examples:

---

Contact

For questions or feedback, feel free to reach out:

• Email: eyal@levys.co.il
• GitHub: https://github.com/elevy99927