LangGraph Multi-Agent DevOps App

This project implements a multi-agent system using LangGraph to create a "DevOps Copilot." The system is composed of a supervisor agent that delegates tasks to specialized agents for planning, diagramming, and Terraform operations.

For a detailed walkthrough of the implementation, check out my Substack post: Building a DevOps Copilot with LangGraph.

Prerequisites

Before you begin, ensure you have the following installed:

• Python 3.10+
• Install GraphViz https://www.graphviz.org/

🚀 Getting Started

Follow these steps to set up and run the project locally.

1. Clone the Repository

git clone git@github.com:blissfulldev/devops-agent.git
cd DevOps-Platform

2. Set Up the Python Environment

# Create and activate a virtual environment
python3 -m venv venv
source venv/bin/activate

# Navigate to the application directory and install dependencies
cd devops-app
pip install poetry
poetry install

3. Configure Environment Variables

Create a .env file from the example template and add your API keys.

# Make sure you are in the `devops-app` directory
cp env.example .env

Now, open the .env file and add the necessary secret keys (e.g., GOOGLE_API_KEY).

▶️ Running the Application

The application consists of several services that must be run simultaneously. It is highly recommended to open a new terminal for each step and activate venv in every terminal before running any command.

1. Start the MCP Servers

These servers provide the specialized tools for each agent.

• Core MCP Server:

  hljs language-bash

  Copy

  cd devops-app/mcp/core-mcp-server/
  poetry install
  

  hljs language-bash

  Copy

  python -m awslabs.core-mcp-server.server --transport streamable-http --host 0.0.0.0 --port 8000
  

• Diagraming MCP Server:

  hljs language-bash

  Copy

  cd devops-app/mcp/aws-diagram-mcp-server/
  poetry install
  

  hljs language-bash

  Copy

  python -m awslabs.aws-diagram-mcp-server.server --transport streamable-http --host 0.0.0.0 --port 8001
  

• Terraform MCP Server:

  hljs language-bash

  Copy

  cd devops-app/mcp/terraform-mcp-server/
  poetry install
  

  hljs language-bash

  Copy

  python -m awslabs.terraform-mcp-server.server --transport streamable-http --host 0.0.0.0 --port 8002
  

2. Start the FastAPI Backend

This server orchestrates the agents and provides the streaming API.

# From the `devops-app` directory
poetry run uvicorn server:app --host 0.0.0.0 --port 8080 --reload

3. Start the Streamlit Frontend

This is the user interface for interacting with the copilot.

# From the `devops-app` directory
streamlit run app.py

You can now access the chat interface at http://localhost:8501.