README.md

Backend – Agentic AI Platform

Overview

This is the FastAPI backend for the Agentic AI Platform. It provides:

• REST API endpoints for LLM (Gemini) and MCP tools
• Configurable, session-based PII masking/restoration
• Dynamic, config-driven pipeline orchestration

---

Setup

1. Install dependencies:

   cd backend
   uv venv
   uv pip install -r requirements.txt

2. Copy and configure environment variables:

   cp .env-sample .env
   # Edit .env with your Gemini API key and config

3. Run the server:

   uvicorn main:app --reload

---

Environment Variables (.env)

• GEMINI_API_KEY – Your Google Gemini API key (required)
• GEMINI_API_URL – (Optional) Override Gemini API endpoint
• GEMINI_MODEL – (Optional) Model name (default: gemini-2.0-flash)
• MCP_CONFIG_PATH – (Optional) Path to MCP config JSON (default: templates/mcp_config.json)

---

Gemini & MCP Config

• Gemini:
  • API key and model are loaded from .env.
  • Endpoint: /llm/gemini (POST)
  • Request: { "prompt": "...", "model": "..." }
  • Response: { "response": "..." }
• MCP Pipeline:
  • Config-driven via mcp_config.json (see templates/)
  • Supports dynamic tool orchestration, pre/post-processing, and session-based PII handling
  • Endpoint: /mcp/pipeline (POST)
  • Tools listed at /mcp/tools

---

Endpoints

• GET / – Root health check
• POST /llm/gemini – Gemini LLM prompt/response
• GET /mcp/health – MCP server health
• GET /mcp/tools – List available MCP tools
• POST /mcp/pipeline – Run pipeline with config-driven tools

---

PII Handling

• In-memory masking/restoration for SSNs, emails, phones, names, addresses, credit cards
• Session-based: Use the same session_id for pre- and post-processing
• Only last 4 digits of credit cards are shown

---

Development Notes

• See main.py for API logic
• See templates/mcp_config.json for pipeline config structure
• Extend MCP tools in mcp_tools/ (if present)