Kube-copilot

Kube-copilot supports the Model Context Protocol (MCP) to allow the LLM to use external tools during diagnose and execute operations.

1. Create an MCP configuration file (e.g., mcp-config.json):

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ]
    },
    "kubernetes": {
      "command": "uvx",
      "args": [
        "mcp-kubernetes-server"
      ]
    }
  }
}

2. Run kube-copilot with the config:

kube-copilot execute --instructions "Your instructions" --mcp-config path/to/mcp-config.json

Kube-copilot supports various LLM providers via environment variables.

OpenAI

Set the OPENAI_API_KEY environment variable.

Anthropic Claude (OpenAI Compatible)

Set the following:

• OPENAI_API_KEY=<your-anthropic-key>
• OPENAI_API_BASE='https://api.anthropic.com/v1/'

Azure OpenAI

Set the following:

• AZURE_OPENAI_API_KEY=<your-api-key>
• AZURE_OPENAI_API_BASE=https://<replace-this>.openai.azure.com/
• AZURE_OPENAI_API_VERSION=2025-03-01-preview

Google Gemini (OpenAI Compatible)

Set the following:

• OPENAI_API_KEY=<your-google-ai-key>
• OPENAI_API_BASE='https://generativelanguage.googleapis.com/v1beta/openai/'

Ollama or other OpenAI compatible LLMs

Set the following:

• OPENAI_API_KEY=<your-api-key>
• OPENAI_API_BASE='http://localhost:11434/v1' (or your own base URL)

Install the kube-copilot CLI using go install. Ensure you have Go installed on your system.

go install github.com/feiskyer/kube-copilot/cmd/kube-copilot@latest

The following flags are available to all kube-copilot subcommands:

• -c, --count-tokens: Print the token count used.
• -x, --max-iterations <int>: Max iterations for the agent running (default: 10 or 30 depending on command context).
• -t, --max-tokens <int>: Max tokens for the GPT model (default: 2048).
• -m, --model <string>: OpenAI model to use (default: gpt-4o).
• -v, --verbose: Enable verbose output.

Use the analyze command to identify potential issues in a specific Kubernetes resource. By default, it analyzes pod resources in the default namespace.

Flags:

• -n, --name <string>: The name of the resource.
• -s, --namespace <string>: The namespace (default: default).
• -r, --resource <string>: The resource type (default: pod).

kube-copilot analyze --name <resource-name> --namespace <namespace> --resource pod

Use the diagnose command to troubleshoot problems occurring within a specific Pod. You can optionally provide an MCP configuration file to enhance diagnosis with external tools.

Flags:

• -n, --name <string>: The name of the Pod.
• -s, --namespace <string>: The namespace (default: default).
• -p, --mcp-config <string>: Path to an MCP configuration file.

kube-copilot diagnose --name <pod-name> --namespace <namespace>

Use the execute command to perform Kubernetes operations or ask questions based on natural language instructions. This command can leverage the Model Context Protocol (MCP) for tool integration.

Flags:

• -i, --instructions <string>: The prompt/instructions to execute.
• -p, --mcp-config <string>: Path to an MCP configuration file.

kube-copilot execute --instructions "Your instructions" --mcp-config path/to/mcp-config.json

Use the generate command to create Kubernetes manifests based on a text prompt. After generation, the CLI will prompt you to confirm whether you want to apply the manifests to your cluster.

Flags:

• -p, --prompt <string>: The prompt used to generate the manifests.

kube-copilot generate --prompt "Create a deployment with 3 replicas of nginx"

Use the audit command to assess security vulnerabilities for a specific Pod. This command requires trivy to be installed on your local machine.

Flags:

• -n, --name <string>: The name of the Pod.
• -s, --namespace <string>: The namespace (default: default).

kube-copilot audit --name <pod-name> --namespace <namespace>

The --prompt flag (or its short form -p) is a required argument for the generate command. It contains the natural language description of the Kubernetes resources you want to create.

-p, --prompt string   Prompts to generate Kubernetes manifests (Required)

The following persistent flags can be used with any kube-copilot subcommand to configure the AI model behavior and execution settings:

--model, -m string
    AI model to use (default: "gpt-4o")

--max-tokens, -t int
    Max tokens for the AI model (default: 2048)

--count-tokens, -c
    Print tokens count

--verbose, -v
    Enable verbose output

--max-iterations, -x int
    Max iterations for the agent running (default: 30)

The following flags are available for the execute command: