mcp-ephemeral-k8s
About
Python implementation to spawn ephemeral Model Context Protocol (MCP) servers using the kubernetes API.
Details
- License
- MIT
Explore
- Supports multiple runtimes:
- Node.js (via npx)
- Python (via uvx)
- Works with mcp-proxy for uvx or npx runtimes
- Supports both local kubeconfig and in-cluster configuration
- Can be run as MCP server
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
mcp-ephemeral-k8sCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"mcp-ephemeral-k8s": {
"mcp-ephemeral-k8s": {
"command": "uvx",
"args": [
"mcp-ephemeral-k8s"
]
}
}
}
}McpServers
{
"mcp-ephemeral-k8s": {
"command": "uvx",
"args": [
"mcp-ephemeral-k8s"
]
}
}
A Python library for spawning ephemeral Model Context Protocol (MCP) servers on Kubernetes using Server-Sent Events (SSE).
- Github: <https://github.com/BobMerkus/mcp-ephemeral-k8s/>
- Documentation: <https://BobMerkus.github.io/mcp-ephemeral-k8s/docs/>
Features
- Supports multiple runtimes:
- Node.js (via npx)
- Python (via uvx)
- Works with mcp-proxy for uvx or npx runtimes
- Supports both local kubeconfig and in-cluster configuration
- Can be run as MCP server
Usage
Running the MCP Server
uvx mcp-ephemeral-k8s
Using the Library
import asyncio
from mcp_ephemeral_k8s import KubernetesSessionManager, presets
async def main():
async with KubernetesSessionManager() as session_manager:
mcp_server = await session_manager.create_mcp_server(
presets.K8S_MCP_SERVER, wait_for_ready=True, expose_port=True
)
print(mcp_server.sse_url)
if __name__ == "__main__":
asyncio.run(main())
Job 'mcp-ephemeral-k8s-proxy-1762291156-x17zuayy' in unknown state, waiting...
http://mcp-ephemeral-k8s-proxy-1762291156-x17zuayy.default.svc.cluster.local:8080/sse
Installation
Prerequisites
- Docker
- Kind or any Kubernetes cluster with valid kubectl configuration
Option 1: Using uvx (Recommended)
uvx mcp-ephemeral-k8s
To connect to the MCP server, use the following config:
{
"mcp-ephemeral-k8s": {
"url": "http://localhost:8000/sse",
"transport": "sse"
}
}
Option 2: As a Python Package
pip install mcp-ephemeral-k8s
mcp-ephemeral-k8s
Option 3: Using Helm Chart
To install the Helm chart, run:helm repo add mcp-ephemeral-k8s https://BobMerkus.github.io/mcp-ephemeral-k8s/
helm repo update
helm install mcp-ephemeral-k8s mcp-ephemeral-k8s/mcp-ephemeral-k8s
To upgrade the Helm chart, run:
helm upgrade -i mcp-ephemeral-k8s mcp-ephemeral-k8s/mcp-ephemeral-k8s
To install a specific version, run:
helm install mcp-ephemeral-k8s mcp-ephemeral-k8s/mcp-ephemeral-k8s --version <replace-with-version>
To uninstall the Helm chart, run:
helm uninstall mcp-ephemeral-k8s
Option 4: From Source
1. Clone the repository
git clone https://github.com/BobMerkus/mcp-ephemeral-k8s.git
cd mcp-ephemeral-k8s
2. Set up development environment
make install
3. Run pre-commit hooks
make check
4. Run tests
make test
5. Build Docker images
make docker-build-local
make docker-build-local-proxy
6. Load images to cluster
kind load docker-image ghcr.io/bobmerkus/mcp-ephemeral-k8s:latest
kind load docker-image ghcr.io/bobmerkus/mcp-ephemeral-k8s-proxy:latest
7. Install Helm chart
helm upgrade -i mcp-ephemeral-k8s charts/mcp-ephemeral-k8s --set image.tag=latest
8. Port forward the MCP server
kubectl port-forward svc/mcp-ephemeral-k8s 8000:8000
9. Visit the FastAPI server
npx @modelcontextprotocol/inspector --sse http://localhost:8000/sse
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
