title	AI Digital Library Assistant
emoji	📚
colorFrom	blue
colorTo	purple
sdk	gradio
sdk_version	5.32.0
app_file	app.py
pinned	false
license	mit
tags	mcp-server-track agent-demo-track llamaindex elevenlabs rag

📚 AI Digital Library Assistant

Your Intelligent Knowledge Companion with Voice & Podcast Capabilities

"Transform your static documents into an interactive, voice-enabled knowledge base that talks back to you."

🔴 LIVE DEMO: Try it on HuggingFace Spaces

---

🌟 What Makes This Special?

The AI Digital Library Assistant is not just another document search tool. It's a comprehensive Model Context Protocol (MCP) server that revolutionizes how you interact with information.

🚀 Killer Features

1. 🎧 World's First Document-to-Podcast Generator: Turn boring reports into engaging, multi-voice conversational podcasts using LlamaIndex analysis and ElevenLabs synthesis.
2. 🎙️ Voice-Enabled Librarian: Talk naturally to your library. The agent understands context, searches documents, and responds with a human-like voice.
3. 🤖 Agentic RAG Workflows: Powered by LlamaIndex ReAct Agents, it doesn't just search—it reasons, plans, and executes complex multi-step research tasks.
4. 🔌 Production-Ready MCP Server: Seamlessly integrate with Claude Desktop or any MCP client to give your LLM direct access to your private knowledge base.

---

⚡ Quick Start

1. Clone the Repository

git clone https://github.com/nihald2000/AI-Digital-Library-Assistant.git
cd AI-Digital-Library-Assistant

2. Install Dependencies

pip install -r requirements.txt

3. Set API Keys

Create a .env file:

OPENAI_API_KEY=sk-...
ELEVENLABS_API_KEY=xi-...
ELEVENLABS_AGENT_ID=... (optional for voice agent)

4. Run the Assistant

python app.py

Access the UI at http://localhost:7860

---

🏗️ System Architecture

The system is built on a modular microservices architecture exposed via the Model Context Protocol.

graph TB
    User[User] --> Gradio[Gradio UI]
    User --> Claude[Claude Desktop]
    
    Gradio --> App[MCP Server]
    Claude --> App
    
    subgraph "Core Services"
        App --> Ingestion[Ingestion Service]
        App --> Search[Search Service]
        App --> RAG[LlamaIndex RAG]
        App --> Voice[ElevenLabs Voice]
        App --> Podcast[Podcast Generator]
    end
    
    subgraph "Data Layer"
        Ingestion --> VectorDB[FAISS Vector Store]
        Ingestion --> DocStore[Document Store]
    end
    
    subgraph "AI Providers"
        RAG --> OpenAI[OpenAI / Mistral]
        Voice --> Eleven[ElevenLabs API]
        Podcast --> Eleven
    end

Loading

---

🛠️ MCP Tools Documentation

The server exposes 8 powerful tools to any MCP-compatible client (like Claude Desktop).

1. ingest_document

Upload and process documents (PDF, TXT, DOCX, Images).

• Args: file_path (str), file_type (str)
• Returns: Document ID and chunk stats.

2. semantic_search

Find relevant content using vector similarity.

• Args: query (str), top_k (int)
• Returns: Ranked list of matching segments.

3. answer_question (RAG)

Ask questions about your documents.

• Args: question (str), context_filter (dict)
• Returns: AI-generated answer with citations.

4. voice_qa 🎙️

Voice-optimized Q&A for conversational interfaces.

• Args: message (str), session_id (str)
• Returns: Speech-ready text response.

5. generate_podcast 🎧

Create an audio podcast from selected documents.

• Args: document_ids (list), style (str), duration_minutes (int)
• Returns: Path to generated MP3 file.

6. summarize_content

Generate summaries in various styles.

• Args: document_id (str), style (str)
• Returns: Summary text.

7. generate_tags

Auto-tag documents for organization.

• Args: document_id (str), max_tags (int)
• Returns: List of tags.

8. list_documents_for_ui

Retrieve library metadata.

• Returns: List of available documents.

---

🎥 Demo Video Script

Watch the full demo here: [YouTube Link Placeholder]

Time	Section	Description
0:00	Intro	"Meet your new AI Digital Library Assistant—it reads, listens, and even makes podcasts."
0:30	Ingestion	Drag-and-drop a complex PDF research paper. Watch it get indexed in seconds.
1:00	Agentic RAG	Ask: "Compare the methodologies in these 3 papers." Watch the agent plan and execute.
2:00	Podcast	THE WOW FACTOR: Select a paper, choose "Conversational" style, and generate a 5-min podcast. Play the audio.
3:30	MCP	Switch to Claude Desktop. Ask Claude to "Read my library and summarize the latest findings."
4:00	Outro	"Built with LlamaIndex, ElevenLabs, and MCP. The future of knowledge management."

---

💻 Technology Stack

We chose the best-in-class tools to build this agent:

Component	Technology	Why We Chose It
Orchestration	LlamaIndex	Best-in-class agentic RAG and data connectors. Essential for complex reasoning.
Voice AI	ElevenLabs	Unmatched voice quality and low latency. Critical for the podcast feature.
Protocol	MCP (FastMCP)	The standard for connecting AI models to data. Enables Claude Desktop integration.
LLM	OpenAI / Mistral	Reliable reasoning and high-context windows for document analysis.
UI	Gradio	Rapid prototyping of beautiful, functional web interfaces.
Vector DB	FAISS	Fast, local, and efficient similarity search.

---

🤝 Sponsor Recognition

A huge thank you to our hackathon sponsors for enabling this innovation:

• LlamaIndex: The ReActAgent and VectorStoreIndex form the brain of our system, allowing it to "think" about documents rather than just search them.
• ElevenLabs: The Text-to-Speech API brings our library to life, powering both the Voice Assistant and the unique Podcast Generator.
• Model Context Protocol: Provided the standardized framework to expose our tools to the world, making this a true "Agentic" platform.

---

🚀 Deployment

Option 1: Local (Recommended for Dev)

python app.py

Option 2: Claude Desktop Integration

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "digital-library": {
      "command": "python",
      "args": ["/absolute/path/to/mcp_server.py"]
    }
  }
}

Option 3: Blaxel Cloud

./deploy.sh

---

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

_{Built with ❤️ for the MCP Hackathon 2024}