Deploying MCP Servers

Deploy a custom MCP server by providing a Docker image that implements the MCP Streamable HTTP Transport protocol.

From the UI

1. Navigate to MCP Servers in the sidebar
2. Click New Server
3. Fill in the form:

Field	Required	Description
Name	Yes	Unique identifier (lowercase, alphanumeric, hyphens)
Image	Yes	Docker image to deploy (e.g., myregistry/my-mcp-server:latest)
Port	No	Container port (default: 8080)
CPU Request	No	CPU cores to request
Memory Request	No	Memory to request

4. Click Deploy

Server URL

The MCP server is accessible at:

https://<domain>/mcp/<name>/

The platform configures a Traefik ingress rule to proxy requests to the server.

Protocol Requirements

Your MCP server must implement the MCP Streamable HTTP Transport specification:

• Accept POST requests at /mcp (or your configured path)
• Handle JSON-RPC 2.0 messages
• Support the MCP initialize, tools/list, and tools/call methods

From the Admin Chat

Use the Admin Chat assistant to deploy servers via natural language:

Deploy a new MCP server called "weather-tools" using image weather-mcp:latest

Environment Variables

To pass environment variables (API keys, config) to your MCP server, edit the Deployment after creation:

kubectl set env deployment/<server-name> \
  API_KEY=your-key \
  -n apps

Health Checks

The platform configures a liveness probe on the server’s HTTP endpoint. If the probe fails repeatedly, Kubernetes restarts the pod and the status changes to Starting.

Deleting an MCP Server

Click the Delete button in the server row. This removes the Deployment, Service, and Ingress rule. No persistent data is deleted (MCP servers are stateless by default).