modules/mcp-server/: Module for deploying MCP servers

Getting Started

Prerequisites

Kubernetes cluster (kind, minikube, or any other Kubernetes cluster)
Terraform (>= 1.3.6) or OpenTofu
kubectl

Note: You need to set the KUBE_CONFIG_PATH environment variable to point to your kubeconfig file:

export KUBE_CONFIG_PATH=${HOME}/.kube/config

This is required by the Helm Terraform provider as the KUBECONFIG environment variable is not respected by this provider. For more information, see the Helm Provider Requirements.

Setup

Create a Kubernetes cluster (if you don't have one):

make cluster

Deploy the kgateway and MCP servers:

make install

Clean up:

make clean

Available Make Targets

  format      - Format all Terraform files in the directory and subdirectories
  lint        - Lint all Terraform files in the directory and subdirectories
  validate    - Validate the Terraform configuration
  checks      - Run format, lint, and validate
  init        - Initialize Terraform configuration
  plan        - Create an execution plan for Terraform changes
  apply       - Apply the Terraform changes
  destroy     - Destroy the Terraform-managed infrastructure
  cluster     - Create a kind cluster named kgateway if it doesn't exist
  install     - Run terraform apply in the terraform directory
  clean       - Delete the kgateway cluster
  port-forward - Forward ports from agent-gateway deployment to localhost
  list-tools  - List available tools from the MCP server
  help        - Show this help message

Using the Model Context Protocol Inspector

The Model Context Protocol (MCP) Inspector is a tool that allows you to interact with MCP servers. This section explains how to use the MCP Inspector to communicate with the agent-gateway.

Local Usage with Port Forwarding

To use the MCP Inspector locally after port forwarding:

Start port forwarding to the agent-gateway:

make port-forward

This command forwards ports 8080 and 19000 from the agent-gateway deployment to your localhost.

In a new terminal, run the MCP Inspector:

# Run the MCP Inspector CLI to list available tools
npx --quiet --no-update-notifier -y @modelcontextprotocol/inspector --cli \
  http://localhost:8080/sse \
  --method tools/list

You can also use the MCP Inspector to call other methods:

# Example: Call a specific tool
npx --quiet --no-update-notifier -y @modelcontextprotocol/inspector --cli \
  http://localhost:8080/sse \
  --method tools/invoke \
  --params '{"name":"example-tool"}'

Using a Pod on the Cluster

You can also run the MCP Inspector from a pod within the Kubernetes cluster:

Fetch the MCP endpoint URL from Terraform/OpenTofu output:

# Using OpenTofu
MCP_ENDPOINT=$(tofu output -raw agent_gateway_mcp_endpoint)

# Or using Terraform
MCP_ENDPOINT=$(terraform output -raw agent_gateway_mcp_endpoint)

echo "MCP endpoint: $MCP_ENDPOINT"

Run the MCP Inspector in a temporary pod:

kubectl run --rm -i --restart=Never --image=docker.io/node:lts-slim mcp-inspector -- \
  /bin/bash -c "npx --quiet --no-update-notifier -y @modelcontextprotocol/inspector --cli $MCP_ENDPOINT --method tools/list"

This is e

Categories

Current Category: kgateway

mcp-dynamic-discovey-poc