RAG_VertexAI_MCP

shilpathota/RAG_VertexAI_MCP

3.1

If you are the rightful owner of RAG_VertexAI_MCP and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

The Enhanced RAG System is a comprehensive solution for chat summarization and document question answering, now upgraded with Model Context Protocol (MCP) for seamless client-backend interaction.

Enhanced RAG System: Chat Summarization + Document QA

Project Overview

This project started as a Chat Summarization API using Vertex AI Gemini to generate concise summaries from chat conversations.

It has now been enhanced to support:

  • Uploading Research Papers / Documents 📄
  • Semantic Search inside documents using FAISS
  • Natural Language Question Answering from uploaded content

✅ Now MCP-Compliant!

This system has been upgraded with MCP (Model Context Protocol) — a standardized interface for unified interaction between the client and backend LLM services.

Built using:

  • FAISS (Facebook AI Similarity Search) for local vector database
  • SentenceTransformer (all-MiniLM-L6-v2) for embeddings
  • Vertex AI Gemini for final answer generation
  • FastAPI for API management
  • Postman for API testing

Features

  • Upload PDF documents.
  • Generate dense embeddings using SentenceTransformer.
  • Store embeddings locally in a FAISS index.
  • Send natural language questions.
  • Retrieve relevant document chunks via semantic search.
  • Generate intelligent, human-like answers using Vertex AI Gemini.

💡 Example Flow (via MCP)

  1. Upload a research paper:

    POST /mcp/upload
    • Form-Data: file=your.pdf
    • Metadata JSON (as metadata field): 
      {
  "request_id": "uuid",
  "context": {
    "type": "upload_doc",
    "data_sources": [],
    "query": "{\"source\": \"chatbot_ui\"}"
  }
}

  2. Ask a question:

    POST /mcp

    {
  "request_id": "uuid",
  "context": {
    "type": "doc_query",
    "data_sources": ["uploaded_docs"],
    "query": "What is the methodology used in this paper?"
  }
}

  3. Summarize a chat:

    POST /mcp

    {
  "request_id": "uuid",
  "context": {
    "type": "summarization",
    "data_sources": [],
    "query": "Customer: Hi, I have an issue...\nAgent: Can you please share your ID?"
  }
}

System Architecture (Block Diagram)

ChatGPT Image Apr 27, 2025, 09_29_35 PM

Updated Architecture with MCP

+-------------+         +---------------------+        +-----------------+
| MCP Client  |  <--->  | FastAPI MCP Server  |  --->  | Vertex AI Gemini |
+-------------+         +---------------------+        +-----------------+
       |                        |                            ↑
       |                        v                            |
       |               Semantic Search (FAISS)              |
       |                        ^                            |
       |                Document Embeddings                 |
       +-----------------+     |     +-----------------------+
                         |     v     |
                         | SentenceTransformer
                         |
                     Uploaded PDFs (pdfplumber)

Tech Stack

  • FastAPI — Lightweight API framework
  • FAISS — Vector similarity search engine
  • SentenceTransformers — For text embeddings
  • Vertex AI Gemini — LLM for content generation
  • pdfplumber — For PDF text extraction
  • Postman — API testing tool

Setup Instructions

  1. Clone the repository
git clone https://github.com/shilpathota/RAGWithVertexAI.git
cd RAGWithVertexAI
  1. Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows use venv\Scripts\activate
  1. Install requirements
pip install -r requirements.txt
  1. Environment Variables

Create a .env file:

GOOGLE_APPLICATION_CREDENTIALS=path/to/your/service-account.json
GOOGLE_PROJECT_ID=your-gcp-project-id
GOOGLE_LOCATION=your-gcp-region
  1. Run the API Server
uvicorn app.main:app --port 8765 --reload

🔌 MCP API Endpoints

1. Unified Request Handling (MCP)

  • POST /mcp
  • Request:
{
  "request_id": "uuid",
  "context": {
    "type": "doc_query" | "summarization",
    "data_sources": [],
    "query": "..."
  }
}

2. Document Upload (MCP)

  • POST /mcp/upload
  • Form-Data:
    • file: PDF file
    • metadata: JSON with context and source

Future Enhancements

  • Document chunking for long PDFs
  • Streaming response from Gemini (for long answers)
  • Real-time upload and QA interface (frontend)
  • Multilingual document support
  • Auto Summarization of full research papers

Author

#AI #MachineLearning #VertexAI #MCP #LLM #FAISS #FastAPI #DocumentSummarization #ChatSummarization #Python #RAG #GitHubProject