Stage 7: Deploy

Duration: ~15 minutes
What happens: AgentSherpa builds a container image, creates an MCPServer CR, and registers your server with the MCP Gateway.

This is the most technical stage. AgentSherpa explains every command before running it.

Step 7a: Build the container image

AgentSherpa uses an OpenShift BuildConfig with the Docker build strategy:

oc new-build --name=${USER_PREFIX}-stock-mcp \
  --strategy=docker \
  --binary

Then starts the build:

oc start-build ${USER_PREFIX}-stock-mcp --from-dir=. --follow

The build runs inside the cluster — no local Docker needed. The resulting image lands in the internal ImageStream.

Step 7b: Get the image reference

After the build completes, the agent retrieves the sha256 image digest:

oc get istag ${USER_PREFIX}-stock-mcp:latest \
  -o jsonpath='{.image.dockerImageReference}'

This digest-pinned reference goes into the MCPServer CR.

Step 7c: Create the MCPServer CR

This is the key moment — instead of manually writing Deployments and Services, you create a single custom resource:

apiVersion: mcp.x-k8s.io/v1alpha1
kind: MCPServer
metadata:
  name: ${USER_PREFIX}-stock-mcp
spec:
  source:
    type: ContainerImage
    containerImage:
      ref: <image-digest-from-7b>
  config:
    port: 8080
    path: /mcp

The MCP Lifecycle Operator automatically creates:

  • A Deployment with security hardening (non-root, drop ALL caps, read-only filesystem, seccomp)

  • A Service (ClusterIP) for internal access

  • A NetworkPolicy for network segmentation

  • An MCP protocol handshake to verify the server responds

AgentSherpa waits for Ready=True:

oc get mcpserver ${USER_PREFIX}-stock-mcp

Expected output:

NAME                   READY   AGE
user1-stock-mcp        True    30s

Step 7d: Verify in AI Hub

Once the MCPServer is Ready, it appears in the OpenShift AI dashboard:

AI Hub → MCP servers → Deployments

You can browse it, see the tools, and verify its status.

Step 7e: Create the HTTPRoute

To expose your server through the MCP Gateway, create an HTTPRoute:

apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: ${USER_PREFIX}-stock-mcp
spec:
  parentRefs:
    - name: mcp-gateway
      namespace: mcp-system
      kind: Gateway
  hostnames:
    - <gateway-hostname>
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /mcp/${USER_PREFIX}-stock
      backendRefs:
        - name: ${USER_PREFIX}-stock-mcp
          port: 8080

The gateway hostname is derived from the cluster route — AgentSherpa calculates it automatically.

Step 7f: Register with the MCP Gateway

The MCPServerRegistration tells the gateway about your tools and adds a prefix to avoid name collisions:

apiVersion: mcp.kuadrant.io/v1alpha1
kind: MCPServerRegistration
metadata:
  name: ${USER_PREFIX}-stock-mcp
spec:
  prefix: ${USER_PREFIX}_stock
  targetRef:
    group: gateway.networking.k8s.io
    kind: HTTPRoute
    name: ${USER_PREFIX}-stock-mcp

With prefix: user1_stock, your tool get_stock_info becomes user1_stock_get_stock_info on the gateway — ensuring no collisions with other participants.

The targetRef points to the HTTPRoute. The gateway discovers the backend service from the route.

Troubleshooting

Issue Check

Build fails

oc logs bc/${USER_PREFIX}-stock-mcp — look for pip install errors or Dockerfile issues

MCPServer stays not-ready

oc describe mcpserver ${USER_PREFIX}-stock-mcp — check events for handshake failures

MCP handshake fails

Verify your server listens on port 8080 at path /mcp and responds to initialize

HTTPRoute not accepted

oc describe httproute ${USER_PREFIX}-stock-mcp — check parentRef matches the gateway

Points

Action Points

MCPServer CR deployed and Ready

15

MCP Gateway registration complete

10

Interactive challenge (post-deploy)

5

Next

Stage 8: Connect & Use