kom is a Kubernetes Operations Manager (MCP server) that exposes a rich set of tools for multi-cluster Kubernetes management. It supports both stdio and SSE modes, enabling lightweight or streaming interactions with Kubernetes resources. With kom, you can perform create/read/update/delete operations on built-in resources and CRDs, manage multiple clusters via RegisterCluster, query resources with SQL-like syntax, retrieve Pod logs, and even perform file operations inside Pods. It also includes capabilities for multi-cluster MCP management, cross-namespace queries, resource event and rollout handling, and integration with MCP tooling through standard endpoints. When running in SSE mode, you can connect to the server’s /sse endpoint to stream updates and responses, which is useful for real-time dashboards and interactive clients. The server is designed to be embedded in MCP tooling and supports integration in workflows that require high-frequency or batch resource operations across clusters.

Prerequisites:

• Access to one or more Kubernetes clusters with a valid kubeconfig.
• Go programming environment (Go 1.18+ recommended) for building from source.

Installation steps:

1. Clone the repository: git clone https://github.com/weibaohui/kom.git cd kom

2. Build the executable:

   From the repository root

   go build main.go

   This will produce a binary named 'kom' (or as configured in your environment)

3. Prepare the runtime environment:

   • Ensure KUBECONFIG is set to your Kubernetes config path, or have in-cluster configuration ready.
   • Optionally configure any required environment variables for authentication or cluster registration as per your deployment environment.

4. Run the server (stdio mode): ./kom

   MCP Server access can be via the default port, typically http://IP:9096/sse for SSE mode or the stdio interface for direct calls.

5. Run the server (SSE mode):

   If you need SSE streaming, ensure the server is started with SSE support and connect clients to /sse

   ./kom

Notes:

• The README indicates the server supports both stdio and SSE modes and can be integrated into MCP tooling via JSON configurations. Adjust the startup flags as needed for your environment.
• When integrating, you may provide the MCP client with the server endpoint (e.g., http://IP:9096/sse) depending on the mode you choose.

Tips and common considerations:

• KUBECONFIG: The server relies on your Kubernetes config to access clusters. Ensure KUBECONFIG points to valid kubeconfig files for the clusters you intend to manage.
• Multi-cluster support: Kom supports multiple clusters and can be registered with different IDs. Use RegisterInCluster or RegisterByPathWithID to add clusters.
• Cross-namespace operations: You can query and operate resources across namespaces, including CRDs, using the provided API methods.
• MCP integration: When wiring into MCP tooling, you can expose the server as an MCP endpoint (stdio or SSE). Example payloads show either a direct command/args entry or an SSE URL for integration.
• Logging and monitoring: Consider enabling Prometheus-style monitoring or logs if your deployment requires observability for resource operations and MCP interactions.
• Environment variables: Typical variables include KUBECONFIG and any cluster-specific auth/config vars required by your deployment. Customize as needed for security and access control.