README.md

---
title: AI Digital Library Assistant
emoji: 📚
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: "5.38.0"
app_file: app.py
pinned: false
license: mit
tags:
  - mcp-in-action-track-consumer
  - mcp-in-action-track-creative
  - building-mcp-track-consumer
  - building-mcp-track-creative
  - MCP-1st-Birthday
---

Demo Link : https://youtu.be/09Lls0zJ-QE

Social media post Link : https://x.com/nihald2000/status/1995198714156286290?s=20

The **AI Digital Library Assistant** is a next-generation knowledge management tool built for the **MCP 1st Birthday Hackathon**. It transforms your static document collection into an interactive, living library.

Unlike traditional RAG (Retrieval Augmented Generation) apps, this project leverages the **Model Context Protocol (MCP)** to create a modular ecosystem of tools—Ingestion, Search, and Podcast Generation—that work harmoniously to help you consume information in the way that suits *you* best.

```mermaid
graph TD
    User((👤 User))
    
    subgraph "Frontend (Gradio)"
        UI[Web Interface]
        PodcastUI[Podcast Studio]
    end
    
    subgraph "MCP Server Layer"
        MCPServer[Content Organizer MCP Server]
        
        subgraph "MCP Tools"
            IngestTool[📥 Ingestion Tool]
            SearchTool[🔍 Search Tool]
            GenTool[✨ Generative Tool]
            PodTool[🎧 Podcast Tool]
        end
    end
    
    subgraph "Service Layer"
        VecStore[(Vector Store)]
        DocStore[(Document Store)]
        LLM[LLM Service (OpenAI / Nebius AI)]
        ElevenLabs[ElevenLabs API]
        LlamaIndex[LlamaIndex Agent]
    end

    User <--> UI
    UI <--> MCPServer
    
    MCPServer --> IngestTool
    MCPServer --> SearchTool
    MCPServer --> GenTool
    MCPServer --> PodTool
    
    IngestTool --> VecStore
    IngestTool --> DocStore
    SearchTool --> VecStore
    GenTool --> LLM
    PodTool --> LlamaIndex
    PodTool --> ElevenLabs
    PodTool --> LLM
```

![AI LIB](https://cdn-uploads.huggingface.co/production/uploads/66f1712d906c08084995f808/TSJexR45eNpUjHhbHDOag.png)


## 🚀 Quick Start

Check out [QUICKSTART.md](QUICKSTART.md) for detailed local setup instructions.

1.  **Clone & Install**:
    ```bash
    git clone https://huggingface.co/spaces/Nihal2000/AiDigitalLibraryAssistant
    pip install -r requirements.txt
    ```
2.  **Configure**: Add your `OPENAI_API_KEY` and `ELEVENLABS_API_KEY` to `.env`.
3.  **Run**: `python app.py`

## 💡 How It Works

### 1. The MCP Core
At the heart of the application is the `AiDigitalLibraryAssistant`. It exposes atomic capabilities (Tools) that the frontend consumes. This means the same tools powering this UI could be connected to Claude Desktop or any other MCP client!
    
    ```bash
    {
      "mcpServers": {
        "ai-library": {
          "command": "npx",
          "args": [
            "-y",
            "mcp-remote",
            "https://mcp-1st-birthday-ai-digital-library-assistant.hf.space/gradio_api/mcp/sse"
          ]
        }
      }
    }
    ```

### 2. 🎧 Podcast Studio (Star Feature)
Turn your reading list into a playlist! The **Podcast Studio** is a flagship feature that transforms any selection of documents into an engaging, multi-speaker audio podcast.
-   **Intelligent Scripting**: Uses **LlamaIndex** and **OpenAI/Nebius AI** to analyze your documents and generate a natural, conversational script.
-   **Multi-Speaker Synthesis**: Leverages **ElevenLabs** to bring the script to life with distinct, realistic voices for each host.
-   **Customizable**: Choose your style (Educational, Casual, Teaching) and duration.


### ✨ Features
## 📚 Document Management

Multi-format Support: PDF, DOCX, TXT, and image files (PNG, JPG, JPEG)
Intelligent OCR: Automatic text extraction from images and scanned documents
Semantic Chunking: Documents automatically split into meaningful segments for better retrieval
Metadata Tracking: Comprehensive document metadata including file size, type, creation date, and custom tags
Vector Embeddings: All documents indexed with dense vector embeddings for semantic search

## 🔍 Advanced Search

Semantic Search: Find documents by meaning, not just keywords
Configurable Results: Adjust the number of results (1-20) based on your needs
Relevance Scoring: Each result includes a confidence score
Source Attribution: Direct links to source documents with highlighted excerpts

## 🎨 Content Studio
Transform your documents with 8 powerful AI tools:

Summarize: Generate concise, detailed, bullet-point, or executive summaries
Generate Outline: Create structured outlines from topics or documents (3-10 sections)
Explain Concept: Get explanations tailored to different audiences (general, technical, beginner, expert)
Paraphrase: Rewrite text in various styles (formal, casual, academic, simple, technical)
Categorize: Automatically classify content into user-defined categories
Key Insights: Extract the most important points from any document
Generate Questions: Create comprehension, analysis, application, creative, or factual questions
Extract Key Info: Pull out structured information (entities, dates, facts) in JSON format

## 🏷️ Smart Tagging

AI-Generated Tags: Automatically generate 3-15 relevant tags for any document
Persistent Storage: Tags saved directly to document metadata
Batch Processing: Tag multiple documents or custom text snippets

## ❓ RAG-Powered Q&A

Context-Aware Answers: Ask questions and get answers grounded in your documents
Source Citations: Every answer includes relevant source excerpts
Confidence Scoring: Transparency about answer reliability
Multi-Document Synthesis: Answers can draw from multiple documents simultaneously

## 🎙️ Podcast Studio
Convert documents into engaging audio conversations:

AI Voice Generation: Ultra-realistic voices powered by ElevenLabs
Two-Host Format: Dynamic dialogue between two AI personalities
Multiple Styles: Conversational, educational, technical, or casual
Custom Duration: 5-30 minute podcasts
Voice Selection: Choose from 7+ professional AI voices
Full Transcripts: Complete text transcripts for every generated podcast
Podcast Library: Browse, play, and manage all generated podcasts

## 📊 Dashboard & Analytics

Real-time Stats: Track total documents, vector chunks, and storage usage
Recent Activity: View recently added documents at a glance
System Health: Monitor vector store, LLM service, and voice service status

## Data Flow

## Document Ingestion: 
    - Files → OCR → Text Extraction → Chunking → Embedding Generation → Vector Store
## Semantic Search: 
    - Query → Embedding → Vector Search → Relevance Ranking → Results
## RAG Q&A: 
    - Question → Search → Context Retrieval → LLM Generation → Answer + Sources
## Podcast Generation: 
    - Documents → Content Analysis → Script Generation → Voice Synthesis → Audio File

### Basic Workflow
1. Upload Documents
Navigate to the "📄 Upload Documents" tab:

Click "Select a document" or drag-and-drop files
Supported formats: PDF, DOCX, TXT, PNG, JPG, JPEG
Click "🚀 Process & Add to Library"
Wait for processing to complete (OCR runs automatically for images)
Note the Document ID from the output

2. Search Your Library
Go to "🔍 Search Documents":

Enter a natural language query (e.g., "What are the key findings about climate change?")
Adjust "Number of Results" slider (1-20)
Click "🔍 Search"
Review results with relevance scores and source excerpts

3. Ask Questions
Navigate to "❓ Ask Questions":

Type your question about uploaded documents
Click "❓ Get Answer"
Receive AI-generated answer with source citations
Check confidence level and review source documents

4. Generate Content
Open "📝 Content Studio":

Select a document from dropdown OR paste custom text
Choose a task from the dropdown:

Summarize, Outline, Explain, Paraphrase, etc.


Configure task-specific options in "⚙️ Advanced Options"
Click "🚀 Run Task"
Copy or download the generated content

5. Create Podcasts
Visit "🎧 Podcast Studio":

Select 1-5 documents using checkboxes
Choose Style (conversational, educational, technical, casual)
Set Duration (5-30 minutes)
Select voices for Host 1 and Host 2
Click "🎙️ Generate Podcast"
Listen to the generated audio and read the transcript
Browse past podcasts in the Podcast Library

6. Generate Tags
Go to "🏷️ Generate Tags":

Select a document OR paste custom text
Adjust "Number of Tags" slider (3-15)
Click "🏷️ Generate Tags"

## 🏆 Hackathon Tracks

We are submitting to:
-   **Building MCP**: For our custom `AiDigitalLibraryAssistant` MCP server implementation.
-   **MCP in Action (Consumer/Creative)**: For the innovative Podcast interface that makes personal knowledge management accessible and fun.

## 📜 License

MIT License. Built with ❤️ for the AI community.

## 🙏 Acknowledgements & Sponsors

This project was built for the **MCP 1st Birthday Hackathon** and proudly leverages technology from:

-   **[OpenAI](https://openai.com)**: Providing the foundational intelligence for our document analysis and content generation.
-   **[Nebius AI](https://nebius.com)**: Powering our high-performance inference needs.
-   **[LlamaIndex](https://www.llamaindex.ai)**: The backbone of our data orchestration, enabling sophisticated RAG and agentic workflows for the Podcast Studio.
-   **[ElevenLabs](https://elevenlabs.io)**: Bringing our podcasts to life with industry-leading, hyper-realistic text-to-speech.
-   **[Hugging Face](https://huggingface.co)**: Hosting our application on **Spaces** and providing the **Gradio** framework for our beautiful, responsive UI.
-   **[Anthropic](https://anthropic.com)**: For pioneering the **Model Context Protocol (MCP)** that makes this modular architecture possible.

## 🔌 Connect to Claude

Want to use these tools directly inside Claude Desktop?
Check out our [Client Setup Guide](CLIENT_SETUP.md) to connect this MCP server to your local Claude instance!