Kubernetes Port Forward MCP

A Model Context Protocol (MCP) server that provides tools for discovering Kubernetes services and running kubectl port-forward sessions (optionally with separate log windows). It is designed for MCP clients/LLMs that translate natural language into structured tool calls.

Kubernetes Port Forward MCP vs kubectl port-forward

This package provides an MCP interface on top of the standard kubectl workflow.

• kubectl: Best when you already know exact namespace/pod/ports and prefer manual control.
• MCP: Best when an agent should discover services and run one or many port-forwards via tool calls.

Key Features

• Service discovery: list namespaces and infer services (short name → environments → namespace) from running pods.
• Multi-service in one session: start multiple port-forwards with one tool call.
• LLM-friendly API: start_k8s_port_forward accepts an array so each service can use different namespace, environment, localPort, and remotePort.
• Optional logs: open kubectl logs -f in a separate OS-level terminal window per service.

Table of Contents

• Requirements
• Quick Start
• Getting Started
• Examples
• Tools
• Output and Logs
• Configuration
• Validation and Debugging

Requirements

• Node.js 18 or newer
• kubectl installed and configured with access to your cluster
• VS Code, Cursor, Windsurf, Claude Desktop, Cline, or any other MCP client

Quick Start

1. Add this server to your MCP client (use the config in Getting Started below).
2. Ask your assistant: "List available Kubernetes services."
3. Ask your assistant: "Run api service on local port 3002."
4. Open the returned URL (for example http://localhost:3002).
5. When finished, ask: "Stop all port-forwards."

Getting Started

First, install the Kubernetes Port Forward MCP server with your client.

Standard config works in most MCP clients:

{
  "mcpServers": {
    "k8s-port-forward": {
      "command": "npx",
      "args": ["-y", "k8s-port-forward-mcp@latest"]
    }
  }
}

"amp.mcpServers": {
  "k8s-port-forward": {
    "command": "npx",
    "args": [
      "-y",
      "k8s-port-forward-mcp@latest"
    ]
  }
}

amp mcp add k8s-port-forward -- npx -y k8s-port-forward-mcp@latest

{
  "mcpServers": {
    "k8s-port-forward": {
      "command": "npx",
      "args": ["-y", "k8s-port-forward-mcp@latest"]
    }
  }
}

claude mcp add k8s-port-forward npx -y k8s-port-forward-mcp@latest

{
  "mcpServers": {
    "k8s-port-forward": {
      "type": "stdio",
      "command": "npx",
      "timeout": 30,
      "args": ["-y", "k8s-port-forward-mcp@latest"],
      "disabled": false
    }
  }
}

codex mcp add k8s-port-forward npx "-y" "k8s-port-forward-mcp@latest"

[mcp_servers.k8s-port-forward]
command = "npx"
args = ["-y", "k8s-port-forward-mcp@latest"]

/mcp add

{
  "mcpServers": {
    "k8s-port-forward": {
      "type": "local",
      "command": "npx",
      "tools": ["*"],
      "args": ["-y", "k8s-port-forward-mcp@latest"]
    }
  }
}

droid mcp add k8s-port-forward "npx -y k8s-port-forward-mcp@latest"

{
  "mcpServers": {
    "k8s-port-forward": {
      "command": "npx",
      "args": ["-y", "k8s-port-forward-mcp@latest"]
    }
  }
}

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "k8s-port-forward": {
      "type": "local",
      "command": ["npx", "-y", "k8s-port-forward-mcp@latest"],
      "enabled": true
    }
  }
}

# For VS Code
code --add-mcp '{"name":"k8s-port-forward","command":"npx","args":["-y","k8s-port-forward-mcp@latest"]}'

{
  "mcpServers": {
    "k8s-port-forward": {
      "command": "npx",
      "args": [
        "-y",
        "k8s-port-forward-mcp@latest"
      ]
    }
  }
}

Examples

Quick reference (natural language → tool calls)

• "Run api and auth services on local ports 3002, 3003"
  → list_k8s_services({}) then start_k8s_port_forward({ services: [{ serviceName: "api", localPort: 3002 }, { serviceName: "auth", localPort: 3003 }] })

• "Run order service from shared services namespace in local port 3000"
  → start_k8s_port_forward({ services: [{ serviceName: "order", namespace: "shared-services", localPort: 3000 }] })

• "Run order service from shared services namespace and remote port 3000 in local port 3000"
  → start_k8s_port_forward({ services: [{ serviceName: "order", namespace: "shared-services", localPort: 3000, remotePort: 3000 }] })

• "Run api service in qa environment on local port 3001"
  → start_k8s_port_forward({ services: [{ serviceName: "api", environment: "qa", localPort: 3001 }] })

• "Run auth service in prod environment from production namespace on local port 3002"
  → start_k8s_port_forward({ services: [{ serviceName: "auth", environment: "prod", namespace: "production", localPort: 3002 }] })

• "Stop all port-forwards"
  → stop_k8s_port_forward({})

Detailed workflow example

User: "Run the frontend service in qa on port 3001"

AI workflow:

1. Call list_k8s_services({}) to find the exact service name.
2. Receive list containing e.g. frontend: qa (ns: ...), dev (ns: ...).
3. Call start_k8s_port_forward({ services: [{ serviceName: "frontend", environment: "qa", localPort: 3001 }] }).
4. Result: port-forwards run in the MCP server process; if includeLogs is true, logs open in a separate window. The tool result includes http://localhost:3001 and the exact kubectl commands.

Tools

Validation and Debugging

Since the MCP server spawns actual kubectl processes in the background, you may want to verify what's running and see the exact commands being executed.

Checking running kubectl processes

Windows (PowerShell):

# List all kubectl processes
tasklist /fi "imagename eq kubectl.exe"

# See the exact commands with process IDs
Get-CimInstance Win32_Process -Filter "Name='kubectl.exe'" | Select-Object ProcessId,CommandLine

Linux/macOS:

# List all kubectl processes
ps aux | grep kubectl

# See the exact commands with process IDs
ps -ef | grep kubectl

This helps you:

• Verify port-forwards are actually running
• See the exact namespaces and ports being used
• Identify stuck processes that might need manual termination
• Debug connectivity issues by examining the actual commands

Common failures and fixes

• Port already in use: choose another local port or stop the conflicting process.
• Connection refused: verify service name, namespace, and selected environment.
• No logs window: set includeLogs: true in the port-forward request.
• Stuck process: terminate by PID (taskkill /PID <pid> on Windows, kill <pid> on Linux/macOS).