Kubernetes MCP Server

K8s MCP Server

This project is a Model Control Protocol (MCP) server for Kubernetes operations. It provides tools for managing deployments, pods, services, and more.

Prerequisites

• Docker
• Python 3.12 or higher
• uv package manager

Building and Running the MCP Server

Using Docker

1. Pull the Docker image:
   docker pull chinnareddy578/k8s-mcp-server
2. Run the Docker container:
   docker run -p 8080:8080 chinnareddy578/k8s-mcp-server
   The server will be accessible at http://localhost:8080.

Using Python Directly

1. Create a virtual environment:
   uv venv .venv
2. Activate the virtual environment:
   • On macOS/Linux:
     source .venv/bin/activate
   • On Windows:
     .venv\Scripts\activate
3. Install dependencies:
   uv pip install -e .
4. Run the server:
   python src/main.py

Using the MCP Server with MCP Clients

GitHub Copilot

1. Configure GitHub Copilot:
   • Open your GitHub Copilot settings.
   • Set the MCP server URL to http://localhost:8080.
2. Use GitHub Copilot:
   • GitHub Copilot will now use the MCP server for code suggestions and completions.

   GitHub Copilot MCP Client Configuration:

   { "servers": { "k8s-mcp-server": { "command": "npx", "args": [ "-y", "mcp-remote", "http://localhost:8080/sse" ] } } }

Claude Desktop

1. Configure Claude Desktop:
   • Open Claude Desktop settings.
   • Set the MCP server URL to http://localhost:8080.
2. Use Claude Desktop:
   • Claude Desktop will now use the MCP server for code suggestions and completions.

   Claude Desktop MCP Client Configuration:

   { "mcpServers": { "k8s-mcp-server": { "command": "npx", "args": [ "-y", "mcp-remote", "http://localhost:8081/sse" ] } } }

Other MCP Clients

For other MCP clients, follow these general steps:

1. Configure the client:
   • Set the MCP server URL to http://localhost:8080.
2. Use the client:
   • The client will now use the MCP server for code suggestions and completions.

Troubleshooting

• Connection Issues: Ensure the Docker container is running and the port is correctly mapped. Check with:
  docker ps
• Tool Not Found: Verify that the tool is registered in your MCP server and the name matches exactly.

License

This project is licensed under the MIT License - see the LICENSE file for details.