MCP servers

Step 1: Deploy an MCP server

Deploy a Model Context Protocol (MCP) server that you want agentgateway to proxy traffic to. The following example sets up a simple MCP server with one tool, fetch, that retrieves the content of a website URL that you pass in.

Notice the following details about the Service:

• appProtocol: agentgateway.dev/mcp (required): Configure your service to use the MCP protocol. This way, the agentgateway proxy uses the MCP protocol when connecting to the service.
• agentgateway.dev/mcp-path annotation (optional): The default values are /sse for the SSE protocol or /mcp for the Streamable HTTP protocol. If you need to change the path of the MCP target endpoint, set this annotation on the Service.

kubectl apply -f- <<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: mcp-website-fetcher
spec:
  selector:
    matchLabels:
      app: mcp-website-fetcher
  template:
    metadata:
      labels:
        app: mcp-website-fetcher
    spec:
      containers:
      - name: mcp-website-fetcher
        image: ghcr.io/peterj/mcp-website-fetcher:main
        imagePullPolicy: Always
---
apiVersion: v1
kind: Service
metadata:
  name: mcp-website-fetcher
  labels:
    app: mcp-website-fetcher
spec:
  selector:
    app: mcp-website-fetcher
  ports:
  - port: 80
    targetPort: 8000
    appProtocol: agentgateway.dev/mcp
EOF

Step 2: Create the backend for the MCP server

Create a AgentgatewayBackend that sets up the agentgateway target details for the MCP server.

kubectl apply -f- <<EOF
apiVersion: agentgateway.dev/v1alpha1
kind: AgentgatewayBackend
metadata:
  name: mcp-backend
spec:
  mcp:
    targets:
    - name: mcp-target
      static:
        host: mcp-website-fetcher.default.svc.cluster.local
        port: 80
        protocol: SSE   
EOF

Step 3: Route to the backend

Create an HTTPRoute resource that routes to the AgentgatewayBackend that you created in the previous step. Use a path match so that requests to /mcp go to the MCP backend and are not routed to an LLM or other backend.

kubectl apply -f- <<EOF
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: mcp
spec:
  parentRefs:
    - name: agentgateway-proxy
      namespace: agentgateway-system
  rules:
    - matches:
      - path:
          type: PathPrefix
          value: /mcp
      backendRefs:
      - name: mcp-backend
        group: agentgateway.dev
        kind: AgentgatewayBackend  
EOF

Step 4: Verify the connection

Use the MCP Inspector tool to verify that you can connect to your sample MCP server through agentgateway.

1. Get the agentgateway address.

   • Cloud Provider LoadBalancer:

     export INGRESS_GW_ADDRESS=$(kubectl get gateway agentgateway-proxy -n agentgateway-system -o=jsonpath="{.status.addresses[0].value}")
     echo $INGRESS_GW_ADDRESS
   • Port-forward for local testing:

     kubectl port-forward deployment/agentgateway-proxy -n agentgateway-system 8080:80

2. From the terminal, run the MCP Inspector command. Then, the MCP Inspector opens in your browser. If the MCP inspector tool does not open automatically, run mcp-inspector.

   npx modelcontextprotocol/inspector#0.18.0

3. From the MCP Inspector menu, connect to your agentgateway address as follows:

   • Transport Type: Select Streamable HTTP.
   • URL: Enter the agentgateway address, port, and the /mcp path. If your agentgateway proxy is exposed with a LoadBalancer server, use http://<lb-address>/mcp. In local test setups where you port-forwarded the agentgateway proxy on your local machine, use http://localhost:8080/mcp.
   • Click Connect.

   

   

4. From the menu bar, click the Tools tab. Then from the Tools pane, click List Tools and select the fetch tool.

5. From the fetch pane, in the url field, enter a website URL, such as https://lipsum.com/, and click Run Tool.

6. Verify that you get back the fetched URL content.