Chapter 3. Deploy and manage MCP servers

Procedure

1. In the Red Hat OpenShift AI dashboard, navigate to AI hub MCP servers.

2. Browse available MCP servers.

   Each server card displays the server name, a description, supported tools, and the support tier. The support tiers are:

   Support tier	Description
   Red Hat	Fully supported by Red Hat.
   Partner	Supported by the technology partner who maintains the server.
   Community	Best-effort support from the open source community.

3. Filter the catalog by deployment mode or supported transport protocols to find the server that meets your requirements.

4. Click the MCP server you want to deploy to open its details page.

   Review the server metadata, including the list of tools it exposes, supported capabilities, and whether the server requires a service account.

5. Click Deploy MCP server.

   Note

   The Deploy MCP server button is disabled if the MCP Lifecycle Operator is not enabled on the cluster. For more information, see Enable the MCP Lifecycle Operator.

6. In the deployment dialog, review the YAML configuration field.

   If the MCP server requires a service account, set the serviceAccountName: field under runtime: and security: in the YAML:

   spec:
     runtime:
       security:
         serviceAccountName: <service_account_name>

   where:

   <service_account_name>

   Specifies the name of the ServiceAccount resource in your namespace.

7. Confirm the deployment by clicking Deploy in the dialog.

   The MCP Catalog submits the deployment request to the MCP Lifecycle Operator. The operator creates an MCPServer custom resource in your namespace and provisions the required Deployment and Service.

Verification

1. Confirm that the MCPServer custom resource shows a READY status of True:

   $ oc get mcpservers -n <namespace>

   Replace <namespace> with your project namespace.

   The output shows the MCP server with READY status as True, for example:

   NAME                 READY   ACCEPTED   IMAGE   PORT   ADDRESS   AGE
   <mcp_server_name>   True    True       ...     ...    ...       2m

2. In the Red Hat OpenShift AI dashboard, navigate to AI hub MCP servers and verify that the MCP server is displayed on the Deployments tab.

Procedure

1. In the Red Hat OpenShift AI dashboard, navigate to AI hub MCP servers and open the Deployments tab.
2. Locate the deployed server and click its name.
3. Click Edit to open the server configuration. Select the new version and save the changes.

Procedure

1. Edit the MCPServer custom resource to update the replica count:

   $ oc patch mcpserver <mcp_server_name> -n <namespace> \
     --type=merge -p '{"spec":{"runtime":{"replicas":<replica_count>}}}'

   where:

   <mcp_server_name>

   Specifies the name of the MCPServer custom resource.

   <namespace>

   Specifies the namespace where the MCP server is deployed.

   <replica_count>

   Specifies the number of replicas to run. Set to 0 to scale to zero.

   You can run oc explain mcpserver.spec.runtime to view all available runtime configuration fields for the installed CRD version.

   Important

   By default, the MCP Lifecycle Operator treats MCP servers as stateful and configures SessionAffinity: ClientIP on the Kubernetes Service. This pins each client to a single pod, which means scaling to additional replicas does not distribute existing client connections across those replicas.

   If the MCP server you are scaling is stateless, which means it does not maintain per-client session data, set spec.mcp.stateless: true in the MCPServer custom resource so that the operator configures SessionAffinity: None and load-balances traffic across all replicas:

   $ oc patch mcpserver <mcp_server_name> -n <namespace> \
     --type=merge -p '{"spec":{"mcp":{"stateless":true},"runtime":{"replicas":<replica_count>}}}'

   Only set stateless: true if the MCP server itself declares that it is stateless.

   The MCP Lifecycle Operator runs MCP servers as Kubernetes-native workloads on OpenShift but does not manage session state within those servers. If the server maintains session state, managing that state across replicas is the server’s own responsibility.

Procedure

1. In the Red Hat OpenShift AI dashboard, navigate to AI hub MCP servers and open the Deployments tab.

2. Locate the MCP server and select Delete from the action menu. Confirm the deletion.

   Alternatively, delete the MCPServer custom resource directly:

   $ oc delete mcpserver <mcp_server_name> -n <namespace>

   The MCP Lifecycle Operator automatically cleans up the associated Deployment, Service, and other managed resources.

Procedure

1. Check the status of all MCP servers in your namespace:

   $ oc get mcpservers -n <namespace>

2. Review the status conditions for a specific MCP server:

   $ oc describe mcpserver <mcp_server_name> -n <namespace>

   Look for a condition with type: Ready and status: True to confirm the server is running and ready to accept connections.