mcp-kubernetes-server

The mcp-kubernetes-server is a server implementing the Model Context Protocol (MCP) to enable AI assistants (such as Claude, Cursor, and GitHub Copilot) to interact with Kubernetes clusters. It acts as a bridge, translating natural language requests from these assistants into Kubernetes operations and returning the results.

It allows AI assistants to:

• Query Kubernetes resources
• Execute kubectl commands
• Manage Kubernetes clusters through natural language interactions
• Diagnose and interpret the states of Kubernetes resources

How It Works

The mcp-kubernetes-server acts as an intermediary between AI assistants (that support the Model Context Protocol) and your Kubernetes cluster. It receives natural language requests from these assistants, translates them into kubectl commands or direct Kubernetes API calls, and executes them against the target cluster. The server then processes the results and returns a structured response, enabling seamless interaction with your Kubernetes environment via the AI assistant.

How To Install

Prerequisites

Before installing mcp-kubernetes-server, ensure you have the following:

• A working Kubernetes cluster.
• A kubeconfig file correctly configured to access your Kubernetes cluster (the server requires this file for interaction).
• The kubectl command-line tool installed and in your system's PATH (used by the server to execute many Kubernetes commands).
• The helm command-line tool installed and in your system's PATH (used by the server for Helm chart operations).
• Python >= 3.11, if you plan to install and run the server directly using uvx (without Docker).

Docker

Get your kubeconfig file for your Kubernetes cluster and setup in the mcpServers (replace src path with your kubeconfig path):

{
  "mcpServers": {
    "kubernetes": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "--mount", "type=bind,src=/home/username/.kube/config,dst=/home/mcp/.kube/config",
        "ghcr.io/feiskyer/mcp-kubernetes-server"
      ]
    }
  }
}

UVX

To run the server using uvx (a tool included with uv, the Python packager), first ensure uv is installed:

Install uv if it's not installed yet and add it to your PATH, e.g. using curl:

# For Linux and MacOS
curl -LsSf https://astral.sh/uv/install.sh | sh

Install kubectl if it's not installed yet and add it to your PATH, e.g.

# For Linux
curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"

# For MacOS
curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/darwin/arm64/kubectl"

Install helm if it's not installed yet and add it to your PATH, e.g.

curl -sSL https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

Config your MCP servers in Claude Desktop, Cursor, ChatGPT Copilot, Github Copilot and other supported AI clients, e.g.

{
  "mcpServers": {
    "kubernetes": {
      "command": "uvx",
      "args": [
        "mcp-kubernetes-server"
      ],
      "env": {
        "KUBECONFIG": "<your-kubeconfig-path>"
      }
    }
  }
}

MCP Server Options

Environment variables:

• KUBECONFIG: Path to your kubeconfig file, e.g. /home/<username>/.kube/config.

Command-line Arguments:

usage: main.py [-h] [--disable-kubectl] [--disable-helm] [--disable-write]
               [--disable-delete] [--transport {stdio,sse,streamable-http}]
               [--host HOST] [--port PORT]

MCP Kubernetes Server

options:
  -h, --help            show this help message and exit
  --disable-kubectl     Disable kubectl command execution
  --disable-helm        Disable helm command execution
  --disable-write       Disable write operations
  --disable-delete      Disable delete operations
  --transport {stdio,sse,streamable-http}
                        Transport mechanism to use (stdio or sse or streamable-http)
  --host HOST           Host to use for sse or streamable-http server
  --port PORT           Port to use for sse or streamable-http server

Usage

Once the mcp-kubernetes-server is installed and configured in your AI client (using the JSON snippets provided in the 'How to install' section for Docker or UVX), you can start interacting with your Kubernetes cluster through natural language. For example, you can ask:

What is the status of my Kubernetes cluster?

What is wrong with my nginx pod?

Verifying the server: If you're running the server with stdio transport (common for uvx direct execution), the AI client will typically start and manage the server process. For sse or streamable-http transports, the server runs independently. You would have started it manually (e.g., uvx mcp-kubernetes-server --transport sse) and should see output in your terminal indicating it's running (e.g., INFO: Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)). You can also check for any error messages in the server terminal if the AI client fails to connect.

Available Tools

The mcp-kubernetes-server provides a comprehensive set of tools for interacting with Kubernetes clusters, categorized by operation type:

Command Tools

These tools provide general command execution capabilities:

Tool	Description	Parameters
kubectl	Run any kubectl command and return the output	command (string)
helm	Run any helm command and return the output	command (string)

Read Tools

These tools provide read-only access to Kubernetes resources:

Tool	Description	Parameters
k8s_get	Fetch any Kubernetes object (or list) as JSON string	resource (string), name (string), namespace (string)
k8s_describe	Show detailed information about a specific resource or group of resources	resource_type (string), name (string, optional), namespace (string, optional), selector (string, optional), all_namespaces (boolean, optional)
k8s_logs	Print the logs for a container in a pod	pod_name (string), container (string, optional), namespace (string, optional), tail (integer, optional), previous (boolean, optional), since (string, optional), timestamps (boolean, optional), follow (boolean, optional)
k8s_events	List events in the cluster	namespace (string, optional), all_namespaces (boolean, optional), field_selector (string, optional), resource_type (string, optional), resource_name (string, optional), sort_by (string, optional), watch (boolean, optional)
k8s_apis	List all available APIs in the Kubernetes cluster	none
k8s_crds	List all Custom Resource Definitions (CRDs) in the Kubernetes cluster	none
k8s_top_nodes	Display resource usage (CPU/memory) of nodes	sort_by (string, optional)
k8s_top_pods	Display resource usage (CPU/memory) of pods	namespace (string, optional), all_namespaces (boolean, optional), sort_by (string, optional), selector (string, optional)
k8s_rollout_status	Get the status of a rollout for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional)
k8s_rollout_history	Get the rollout history for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional), revision (string, optional)
k8s_auth_can_i	Check whether an action is allowed	verb (string), resource (string), subresource (string, optional), namespace (string, optional), name (string, optional)
k8s_auth_whoami	Show the subject that you are currently authenticated as	none

Write Tools

These tools provide create, update or patch operations to Kubernetes resources:

Tool	Description	Parameters
k8s_create	Create a Kubernetes resource from YAML/JSON content	yaml_content (string), namespace (string, optional)
k8s_apply	Apply a configuration to a resource by filename or stdin	yaml_content (string), namespace (string, optional)
k8s_expose	Expose a resource as a new Kubernetes service	resource_type (string), name (string), port (integer), target_port (integer, optional), namespace (string, optional), protocol (string, optional), service_name (string, optional), labels (object, optional), selector (string, optional), type (string, optional)
k8s_run	Create and run a particular image in a pod	name (string), image (string), namespace (string, optional), command (array, optional), env (object, optional), labels (object, optional), restart (string, optional)
k8s_set_resources	Set resource limits and requests for containers	resource_type (string), resource_name (string), namespace (string, optional), containers (array, optional), limits (object, optional), requests (object, optional)
k8s_set_image	Set the image for a container	resource_type (string), resource_name (string), container (string), image (string), namespace (string, optional)
k8s_set_env	Set environment variables for a container	resource_type (string), resource_name (string), container (string), env_dict (object), namespace (string, optional)
k8s_rollout_undo	Undo a rollout for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional), to_revision (string, optional)
k8s_rollout_restart	Restart a rollout for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional)
k8s_rollout_pause	Pause a rollout for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional)
k8s_rollout_resume	Resume a rollout for a deployment, daemonset, or statefulset	resource_type (string), name (string), namespace (string, optional)
k8s_scale	Scale a resource	resource_type (string), name (string), replicas (integer), namespace (string, optional)
k8s_autoscale	Autoscale a deployment, replica set, stateful set, or replication controller	resource_type (string), name (string), min (integer), max (integer), namespace (string, optional), cpu_percent (integer, optional)
k8s_cordon	Mark a node as unschedulable	node_name (string)
k8s_uncordon	Mark a node as schedulable	node_name (string)
k8s_drain	Drain a node in preparation for maintenance	node_name (string), force (boolean, optional), ignore_daemonsets (boolean, optional), delete_local_data (boolean, optional), timeout (integer, optional)
k8s_taint	Update the taints on one or more nodes	node_name (string), key (string), value (string, optional), effect (string)
k8s_untaint	Remove the taints from a node	node_name (string), key (string), effect (string, optional)
k8s_exec_command	Execute a command in a container	pod_name (string), command (string), container (string, optional), namespace (string, optional), stdin (boolean, optional), tty (boolean, optional), timeout (integer, optional)
k8s_port_forward	Forward one or more local ports to a pod	resource_type (string), name (string), ports (array), namespace (string, optional), address (string, optional)
k8s_cp	Copy files and directories to and from containers	src_path (string), dst_path (string), container (string, optional), namespace (string, optional)
k8s_patch	Update fields of a resource	resource_type (string), name (string), patch (object), namespace (string, optional)
k8s_label	Update the labels on a resource	resource_type (string), name (string), labels (object), namespace (string, optional), overwrite (boolean, optional)
k8s_annotate	Update the annotations on a resource	resource_type (string), name (string), annotations (object), namespace (string, optional), overwrite (boolean, optional)

Delete Tools

These tools provide delete operations to Kubernetes resources:

Tool	Description	Parameters
k8s_delete	Delete resources by name, label selector, or all resources in a namespace	resource_type (string), name (string, optional), namespace (string, optional), label_selector (string, optional), all_namespaces (boolean, optional), force (boolean, optional), grace_period (integer, optional)

Development

How to run the project locally:

uv run -m src.mcp_kubernetes_server.main

How to inspect MCP server requests and responses:

npx @modelcontextprotocol/inspector uv run -m src.mcp_kubernetes_server.main

Troubleshooting

Here are some common issues and their solutions when working with mcp-kubernetes-server:

Issue: mcp-kubernetes-server cannot connect to the Kubernetes cluster or reports authentication errors. Solution:

• Ensure your kubeconfig file is correctly configured and points to the intended cluster.
• Verify that the path to your kubeconfig file is correctly specified in the mcpServers configuration (for Docker, ensure the mount path is correct; for uvx, ensure the KUBECONFIG environment variable is set correctly).
• Check that the credentials in your kubeconfig have the necessary permissions to perform operations on the cluster. You can test this with kubectl directly (e.g., kubectl get pods).

Issue: kubectl or helm commands return an error like "command not found" or are disabled. Solution:

• If running via uvx, ensure kubectl and/or helm are installed on your system and available in your PATH. Refer to the "Prerequisites" section for installation guidance.
• If you see a message like "Write operations are not allowed" or "Delete operations are not allowed", the server might have been started with flags like --disable-kubectl, --disable-helm, --disable-write, or --disable-delete. Check the server's startup command and the "MCP Server Options" section in the README for details on these flags.

Issue: How can I see the raw requests and responses between my AI client and the mcp-kubernetes-server? Solution:

• You can use the @modelcontextprotocol/inspector tool as mentioned in the "Development" section: npx @modelcontextprotocol/inspector uv run -m src.mcp_kubernetes_server.main. This will show you the MCP messages being exchanged.

Issue: The server starts but the AI client cannot connect. Solution:

• If using stdio transport (default for uvx direct execution), ensure your AI client is configured to launch the mcp-kubernetes-server command correctly.
• If using sse or streamable-http transport, ensure the host and port configured in the mcp-kubernetes-server (e.g., --host 0.0.0.0 --port 8000) are reachable from where your AI client is running. Check for firewall rules or network configuration issues. Also, verify the AI client is configured with the correct URL for the server.

Contribution

This project is open source, available on GitHub at feiskyer/mcp-kubernetes-server and licensed under the Apache License.

If you would like to contribute to the project, please follow these guidelines:

1. Fork the repository and clone it to your local machine.
2. Create a new branch for your changes.
3. Make your changes and commit them with a descriptive commit message.
4. Push your changes to your forked repository.
5. Open a pull request to the main repository.

LICENSE

The project is licensed under the Apache License 2.0. See the LICENSE file for more details.