sawnjordan/k8s-agent-mcp-server-langgraph

Kubernetes & S3 MCP Servers with LangGraph + Streamlit

This project implements a multi-server MCP (Multi-Server Chat Protocol) setup integrating:

🧩 LangGraph for multi-agent orchestration

🤖 DeepSeek or MistralAI as the LLM backend

☸️ Kubernetes MCP Server for managing and querying clusters

🪣 S3 MCP Server for S3-based data management

💬 Streamlit UI for interactive conversation with both servers

Prerequisites

Before you start, ensure the following requirements are met.

System Requirements

• Platform: Linux (recommended) or macOS

• Python ≥ 3.10

• Docker and Docker Compose v2+ installed

• kubectl CLI configured and working

• Git installed

For Kubernetes MCP Server

Depending on your environment:

🧱 Local (Kind) Cluster

• You must have Kind installed and a cluster created:

kind get clusters

• Verify connectivity:

kubectl get nodes

• Your local kubeconfig (~/.kube/config) must be accessible.

☁️ Cloud Cluster (EKS / GKE / AKS)

• Configure your cluster credentials locally( Any cloud ). Below one is for AWS.

aws eks update-kubeconfig --name <cluster-name> --region <region> --profile <aws-profile>

• Confirm access:

kubectl get pods -A

• 🧩 The MCP Server reads your kubeconfig inside the container.

🪣 For S3 MCP Server

To enable the S3 integration:

• Create a .env.aws file (based on .env.aws.example):

AWS_ACCESS_KEY_ID=<your-access-key>
AWS_SECRET_ACCESS_KEY=<your-secret-key>
AWS_DEFAULT_REGION=<your-region>
AWS_SESSION_TOKEN=<your-session-token>
AWS_PROFILE=<your-profile>

These credentials must have s3:* and (optionally) kms:* permissions.

• This file is required for S3 MCP to authenticate successfully.

• If using AWS SSO instead of static keys, make sure your host machine has valid SSO config under ~/.aws/config

You can verify it with:

aws sts get-caller-identity --profile <profile-name>

Environment Setup

1. Clone the repository:

git clone https://github.com/sanjog-lama/k8s-agent-mcp-server-langgraph.git
cd k8s-agent-mcp-server-langgraph

2. Create and activate a virtual environment:

python3 -m venv venv
source venv/bin/activate

3. Install required packages:

pip install -r requirements.txt

4. Create a .env file in the project root and add your API keys:

MISTRAL_API_KEY=your_mistral_key
OR
DEEPSEEK_API_KEY=your_mistral_key

3. Running the Project:

Open two separate terminals:

• Terminal 1 - Custom MCP Server - Start the MCP server:

python k8_mcp_server.py

This will run your Kubernetes MCP server and expose the tools via streamable-http transport.

• Terminal 2 - MCP Client

You can interact with the MCP server in two ways:

• Option 1: Run the client script

python mcp_client_langgraph.py

This runs a CLI-based interaction using LangGraph.

• Option 2: Run the Streamlit web app

streamlit run web_app.py

Sample Prompt

List pods on <custom-name> namespace

Using Docker for build and run

If you are using Kind Cluster on local

Note: If you are using other local cluster then kind no need to do below:

• Create .env.kind file as shown on .env.kind.sample.
• Run below commands:

chmod +x ./get_kind_control_plane.sh
chmod +x ./build_and_run_kind.sh

• To get the kind cluster control plane IP and Port run below command:

./get_kind_control_plane.sh

Note: You can copy the output on .env.kind.

• To build and run the MCP servers and web UI.

./build_and_run_kind.sh

If you are using Cluster on Cloud (AWS)

• Create .env.aws file with the AWS credentials as shown on .env.aws.example. This is required for s3 mcp server.
• Run below command

chmod +x ./build_and_run.sh
./build_and_run.sh

Chat History

• Chat history are saved on ./data/.