TahreemShahid/MCP_Chatbot — reverse-engineered prompt

Build me a full stack document chat app where I can upload a PDF and then ask questions about it in a chat window.

I want the app to read the PDF, break it into searchable parts, and answer based on the document instead of guessing. The chat should feel live, with the answer streaming in as it’s written. Each uploaded PDF should have its own separate chat session, and I should be able to clear or remove a session.

Use a Python FastAPI backend and a React frontend. The backend should use an agent that can decide when to search the uploaded document, with real MCP tool integration for document search, FAISS for the local vector search, Gemini embeddings, and Groq for the chat model. Please include environment variable setup for the needed API keys.

Also include a simple evaluation script that uploads a test PDF, asks a few known questions, and reports accuracy and response times. Make it easy to run locally in development and also serve the built frontend from the backend for production.