Minimalist SRE-Agent Playground with MCP Tools

Prerequisites

To use the SRE-Agent playground, you need the following installed or running in your OS:

• access to a Kubernetes cluster with KUBECONFIG environment variable pointing to the cluster's kubeconfig file.

  # add in your .bashrc/.bash_profile or .zshrc/.zshenv
  export KUBECONFIG=/path/to/kubeconfig_file

• uv command

    # For Linux and MacOS
    curl -LsSf https://astral.sh/uv/install.sh | sh

• kubectl command

    # For Linux
    curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
    # For MacOS
    curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/darwin/arm64/kubectl"

• helm command

    curl -sSL https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

• node command

  # For MacOS
  # using brew
  brew install node
  
  # For Linux/MacOS
  # using curl
  # Download and install nvm:
  curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
  # restart the shell
  nvm install 25

• ollama server for serving local agents

  # install ollama in linux
  curl -fsSL https://ollama.com/install.sh | sh
  
  # install ollama using brew in MacOS
  brew install ollama
  # or download the ollama application
  curl -OL "https://ollama.com/download/Ollama.dmg"

• qwen3:8b local llm running in ollama server

  # install qwen3 chatbot
  ollama pull qwen3:8b
  ollama list
  # test the chatbot
  ollama run qwen3:8b

• Flowise GUI framework for Agentflow or Chatflow

  # install flowise
  npm install -g flowise
  # start flowise and access it at http://localhost:3000
  npx flowise start

Flowise Framework for AI Applications

There are three major options to build AI-powered applications in Flowise, namely: Assistants, Chatflows, and Agentflows. An AI Assistant provides the simplest way to build AI applications using preconfigured structure for rapid prototyping. It lacks customization but it can be a good choice to try and explore a running application in the flowise framework.

Chatflow focuses on a single agent doing tasks using different sources of information while Agentflow covers advanced features for multi-steps reasoning, multi-agents interaction, decision-making, planning, orchestration, and context engineering. Since Agentflow is a superset of Chatflow and provides the most customization among the three, its framework will be used in the implementation of a Simple SRE Agent and its customizations and configurations are listed in detail below.

Simple Agent Mode

• Start flowise if it is not running: npx flowise start

• Open flowise in a browser at http://localhost:3000

• Navigate at the left-hand menu and click Agentflows to open the main panel and click the Add New button located at the upper right corner.

• Click the plus button to open the available nodes. Drag-and-drop the Agent node.

• Connect the start node to the Agent node.

• Double-click the Agent node to open the custimization options. In the Model section, search for ChatOllama.

• Click the ChatOllama Parameters drop-down menu. The Base URL is already correct. Add qwen3:8b in the Model Name. Leave the other options in their default values.

• Goto the Tools section. Search for the keyword MCP and choose Custom MCP.

• Click the Custom MCP Parameters menu. Copy and paste the following json configuration:

  {
    "command": "uvx",
    "args": [
        "--from",
        "git+https://github.com/mdwoicke/kuber-mcp-server.git",
        "kubectl-mcp"
    ],
    "env": {
        "KUBECONFIG": "/absolute/path/to/.kube/config"
  }
  

  

• Click the refresh tools and select the actions from the drop-down menu.

• As you choose the actions in the list, the tool list will be updated.

• Finally, look for the Messages section and add a system prompt.

  

• You can close the Agent customization options by clicking anywhere outside the customization box. At the upper-right corner, click the chat icon to open the chat interface as shown below. Type your query in the designated chatbox.

  

  NOTE: One can skip the manual creation of this simple SRE agent by loading the json file called simple-sre-agent.json in the agents directory. Navigate at the left-hand menu and click Agentflows to open the main panel and click the Add New button located at the upper right corner. Once the Agentflow panel is active, click the gear icon at the upper right corner and choose Load Agents. Clone the repo so that you have the copy of the json file that you can load. Just remember to edit the Custom MCP Parameters of the agent with the following content:

{
  "command": "uvx",
  "args": [
      "--from",
      "git+https://github.com/mdwoicke/kuber-mcp-server.git",
      "kubectl-mcp"
  ],
  "env": {
      "KUBECONFIG": "/absolute/path/to/.kube/config"
}