Infra Pilot: Terraform MCP Server

Infra Pilot is an MCP-compliant, agentic DevOps copilot that understands and answers questions about your Terraform infrastructure code using a local LLM (LLaMA via Ollama), LangGraph agent flow, and vector-based RAG.

Think of it as a ChatGPT-like assistant that knows your actual Terraform setup — and can generate documentation, answer questions, and be extended to detect drift, security issues, or cost spikes.

---

Features

• Local LLM inference (LLaMA 3 via Ollama — no OpenAI key required)
• LangGraph-powered multi-step agent with structured context
• Embeds .tf files and answers natural language questions
• Auto-generates Markdown documentation from code
• Streamlit UI for easy querying
• FastAPI backend with /ask and /docs/generate

---

Project Structure

infra-pilot/
├── app/
│   ├── main.py                # FastAPI app
│   ├── routes/                # /ask and /docs endpoints
│   ├── services/              # Embedding + inference logic
│   ├── agents/                # LangGraph reasoning flow
│   ├── templates/             # Structured prompt template
│   └── ui/app.py              # Streamlit UI
├── infra/                    # Your Terraform code (.tf files)
├── chroma_store/             # Vector DB store (auto-generated)
├── docs_output/              # Markdown docs (generated)
├── embed_terraform.py        # Script to re-embed .tf files
├── requirements.txt
├── setup.sh                  # Quick start setup script
└── README.md

---

How It Works (MCP Pipeline)

1. Code context is embedded with HuggingFace + Chroma
2. A LangGraph agent uses a structured prompt template:

## Code: {code_chunks}
## Vars: {tf_vars}
## Question: {user_query}

3. LLaMA model (via Ollama) generates answers
4. Markdown docs are generated with the same reasoning chain

---

Setup Instructions

1. Run the Setup Script

chmod +x setup.sh
./setup.sh

This will:

• Create a virtual environment
• Install all dependencies
• Prompt you to run embedding and start the app

2. Or Follow Manual Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
ollama run llama3
python embed_terraform.py
uvicorn app.main:app --reload
streamlit run app/ui/app.py

---

Example Questions (Try These in Streamlit)

• "What does this module create?"
• "What IAM permissions are given to Lambda?"
• "Is the RDS database publicly accessible?"
• "List the input variables and their defaults."

---

Generate Docs

Click " Generate Markdown Documentation" in the Streamlit UI, or call it directly:

curl -X POST http://localhost:8000/docs/generate

Output saved to:

docs_output/infra-doc-YYYYMMDD-HHMMSS.md

---

Roadmap Ideas

• Slack bot interface: /ask-infra
• Drift detection with AWS SDK / TF state
• Cost estimation with Infracost
• Test mode for mocking plan/apply
• Auto-export docs to GitHub/Confluence

Contributing

Want to build a plugin, add an agent tool, or enhance UI? PRs and ideas welcome — let’s build InfraPilot into the ultimate infra copilot.

---

Questions / Feedback

Open an issue or ping @siva if you need help, extensions, or a demo setup!