Kubernetes MCP Server (FastMCP)

This project provides a Kubernetes MCP (Model Context Protocol) server built using FastMCP and the official Kubernetes Python client. It exposes a wide range of Kubernetes management and inspection tools as MCP tools, making it easy to interact with your Kubernetes or OpenShift cluster programmatically or via GitHub Copilot.

use https://gitingest.com/jlowin/fastmcp

download the above and feed it as a context to your copilot chat

prompt to build a mcp server: build a kubernet mcp server with top 30 kubernetes tools using the jlowin-fastmcp-8a5edab282632443.txt file. Make the functions dynamic so that it works for all cases even when the user does not provide a namespace or resource name.

Features

The MCP server exposes the following Kubernetes operations as MCP tools:

• Pod Management: List, describe, and get logs for pods
• Deployment Management: List, scale, and inspect deployments
• Service Management: List and get details of services
• Namespace Management: List, create, and delete namespaces
• Node Management: List and describe nodes
• ConfigMap & Secret Management: List and get ConfigMaps and Secrets
• Persistent Volumes & Claims: List PVs and PVCs, get PVC details
• Job & CronJob Management: List and get jobs and cronjobs
• Ingress Management: List and get ingresses
• DaemonSet, StatefulSet, ReplicaSet Management: List and get details for each
• Event Management: List and get cluster events

All tools support filtering by namespace where applicable.

Prerequisites

• Python 3.x
• fastmcp
• kubernetes Python client
• kubectl (for cluster access)
• Access to a Kubernetes or OpenShift cluster (kubeconfig or in-cluster config)

Installation

1. Install Python dependencies:

pip install fastmcp kubernetes

2. Ensure `python` and `fastmcp` are in your system PATH.

3. Install and configure `kubectl` and connect to your cluster.

4. (Optional) Update your `mcp.json` to add the MCP server endpoint:

"kubernetes-local": {
      "name": "Kubernetes MCP Server",
      "url": "http://localhost:8000/mcp/"
    }

Usage

1. Start the MCP server:

python kube_mcp_server.py

2. The server will run on `http://0.0.0.0:8000/mcp/` by default.

3. You can now interact with your Kubernetes cluster using MCP tools (e.g., via GitHub Copilot or any MCP client).

Example MCP Tools

• `list_pods(namespace=None)`
• `get_pod_logs(pod_name=None, namespace=None)`
• `describe_pod(pod_name=None, namespace=None)`
• `list_deployments(namespace=None)`
• `scale_deployment(deployment_name, replicas, namespace)`
• `list_services(namespace=None)`
• `get_service(service_name=None, namespace=None)`
• `list_namespaces()`
• `create_namespace(namespace)`
• `delete_namespace(namespace)`
• `list_nodes()`
• `describe_node(node_name=None)`
• `list_configmaps(namespace=None)`
• `get_configmap(configmap_name=None, namespace=None)`
• `list_secrets(namespace=None)`
• `get_secret(secret_name=None, namespace=None)`
• `list_persistent_volumes()`
• `list_persistent_volume_claims(namespace=None)`
• `get_pvc(pvc_name=None, namespace=None)`
• `list_jobs(namespace=None)`
• `get_job(job_name=None, namespace=None)`
• `list_cronjobs(namespace=None)`
• `get_cronjob(cronjob_name=None, namespace=None)`
• `list_ingresses(namespace=None)`
• `get_ingress(ingress_name=None, namespace=None)`
• `list_daemonsets(namespace=None)`
• `get_daemonset(daemonset_name=None, namespace=None)`
• `list_statefulsets(namespace=None)`
• `get_statefulset(statefulset_name=None, namespace=None)`
• `list_events(namespace=None)`
• `get_event(event_name=None, namespace=None)`
• `list_replicasets(namespace=None)`
• `get_replicaset(replicaset_name=None, namespace=None)`

Notes

Example prompt:

1. get me the pods running on the vp1099 namespace.

2. scale down all pods in the vp1099 namespace to 1 replica.

• The server will attempt to use your local kubeconfig, or fall back to in-cluster config if running inside a cluster.
• Make sure your `kubectl` context is set to the desired cluster/namespace.
• You can extend the server by adding more MCP tools using the `@mcp.tool` decorator.
• Gitingest support: You can use gitingest to ingest and interact with GitHub repositories directly through MCP, enabling advanced code search and automation workflows alongside your Kubernetes operations.

License

MIT License