MCP Setup

CLI Plugin Setup

This skill helps you install the oc plugins required by the other skills in this collection.

What to Check

First, check if the plugins are already installed:

oc mtv --help 2>/dev/null && echo "MTV_OK" || echo "MTV_MISSING"
oc metrics --help 2>/dev/null && echo "METRICS_OK" || echo "METRICS_MISSING"
oc debug-queries --help 2>/dev/null && echo "DEBUG_OK" || echo "DEBUG_MISSING"

How to Respond

Based on the results above, tell the user only what is missing and provide the relevant install instructions. If everything is already installed, confirm it and move on.

Install all tools at once

Use the secure version-pinned installer (downloads binaries, verifies SHA256 checksums):

curl -sSLO https://raw.githubusercontent.com/kubev2v/mtv-skills/main/tools/install-tools.sh
curl -sSL  https://raw.githubusercontent.com/kubev2v/mtv-skills/main/SHA256SUMS | shasum -a 256 --check --ignore-missing
bash install-tools.sh kubectl-mtv kubectl-metrics kubectl-debug-queries && rm install-tools.sh

Or install individually:

bash install-tools.sh kubectl-mtv
bash install-tools.sh kubectl-metrics
bash install-tools.sh kubectl-debug-queries

Verify:

oc mtv --help
oc metrics --help
oc debug-queries --help

Versions and SHA256 checksums are pinned in tools/versions.json. To update to a new release, maintainers run tools/update-versions.sh.

PATH setup

All three install to ~/.local/bin by default. If it is not in the user's PATH:

# bash
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc

# zsh
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc

Tool Summary

Note: kubectl mtv, kubectl metrics, and kubectl debug-queries also work as aliases.

Optional: MCP Server Configuration

If you are using an AI agent that supports MCP (Model Context Protocol), these plugins can also run as MCP servers. This is optional — the skills work with the CLI directly.

Claude Code (CLI)

claude mcp add kubectl-metrics -- oc metrics mcp-server
claude mcp add kubectl-mtv -- oc mtv mcp-server
claude mcp add kubectl-debug-queries -- oc debug-queries mcp-server

Cursor IDE

Settings -> MCP -> Add Server for each:

Claude Desktop

Edit claude_desktop_config.json and add to the mcpServers section:

{
  "mcpServers": {
    "kubectl-metrics": {
      "command": "oc",
      "args": ["metrics", "mcp-server"]
    },
    "kubectl-mtv": {
      "command": "oc",
      "args": ["mtv", "mcp-server"]
    },
    "kubectl-debug-queries": {
      "command": "oc",
      "args": ["debug-queries", "mcp-server"]
    }
  }
}

SSE Mode (OpenShift Lightspeed or Remote Agents)

For remote or server-based agents, run each MCP server in SSE mode:

oc metrics mcp-server --sse --port 8080
oc mtv mcp-server --sse --port 8081
oc debug-queries mcp-server --sse --port 8082

Container Images (no local binary needed)

Run MCP servers as containers instead of installing the plugins locally. Requires Docker or Podman and a valid cluster token:

# kubectl-mtv
docker run --rm -p 8080:8080 \
  -e MCP_KUBE_SERVER=https://api.cluster.example.com:6443 \
  -e MCP_KUBE_TOKEN=sha256~xxxx \
  quay.io/yaacov/kubectl-mtv-mcp-server:v0.3.26

# kubectl-metrics
docker run --rm -p 8081:8080 \
  -e MCP_KUBE_SERVER=https://api.cluster.example.com:6443 \
  -e MCP_KUBE_TOKEN=sha256~xxxx \
  quay.io/yaacov/kubectl-metrics-mcp-server:v0.1.13

# kubectl-debug-queries
docker run --rm -p 8082:8080 \
  -e MCP_KUBE_SERVER=https://api.cluster.example.com:6443 \
  -e MCP_KUBE_TOKEN=sha256~xxxx \
  quay.io/yaacov/kubectl-debug-queries-mcp-server:v0.1.5

Then configure your agent to connect via SSE at http://localhost:8080/sse, http://localhost:8081/sse, and http://localhost:8082/sse.

Deploy on OpenShift

Deploy the MCP servers directly on the cluster (pin to the same versions as in tools/versions.json):

# kubectl-mtv
oc apply -f https://raw.githubusercontent.com/yaacov/kubectl-mtv/v0.3.26/deploy/mcp-server.yaml

# kubectl-debug-queries
oc apply -f https://raw.githubusercontent.com/yaacov/kubectl-debug-queries/v0.1.5/deploy/mcp-server.yaml