Fuel MCP Server

🚀 Fuel Network Document Retrieval System

This project offers a comprehensive document retrieval solution based on the QdrantDB vector database and the MCP (Model Context Protocol). By integrating a locally running Qdrant instance, users can efficiently search and query technical document content. The system includes multiple functional modules such as document indexing, vector embedding generation, and integration with IDEs supporting the MCP protocol.

🚀 Quick Start

✨ Features

• Document Indexer: Supports Markdown format, automatically identifies and extracts structured content such as code blocks and paragraphs. Performs intelligent text chunking based on the target word count to ensure content integrity and retrieval efficiency. Converts text into high - dimensional vector representations using the Transformers.js library.
• Retrieval Engine: Uses cosine similarity for vector comparison and returns the most relevant document fragments. Dynamically sorts retrieval results based on the matching degree. Improves the relevance of retrieval results based on the information of adjacent chunks.
• MCP Protocol Integration: Seamlessly integrates with development environments supporting the MCP protocol, such as VS Code and IntelliJ IDEA. Allows users to trigger document retrieval functions directly during code editing to quickly obtain required technical information.
• Scalability Design: Plans to add support for content types such as images and audio in the future. Is planning a microservice - based deployment solution using Kubernetes. Is developing real - time file change detection and automatic index update functions.

📦 Installation

Dependencies

• Node.js: Version v16.0.0 or higher is required.
• Docker: Used to run the QdrantDB container.
• Python Transformers Library: Used for text embedding generation.
• File Watchers: Optional, used for automatically indexing file changes.

Necessary Environment Variables

# Qdrant database configuration
export QDRANT_URL=http://localhost:6333  # Default local running address
export QDRANT_API_KEY=your_api_key       # Fill in according to the actual API key

# Model - related configuration
export EMBEDDING_MODEL=Xenova/all - MiniLM - L6 - v2  # Default text embedding model

💻 Usage Examples

Basic Usage

# Initialize the project
git clone [your - repository - url]
cd fuel - network - document - system
npm install

# Start the QdrantDB service
docker run -p 6333:6333 -p 6334:6334 \
    -v $(pwd)/qdrant_storage:/qdrant/storage \
    qdrant/qdrant

# Start the document indexing service
rm -rf qdrant_storage  # Optional: Delete the old database
npm run dev

Advanced Usage

# Document Indexing
bun run indexer -- docsDir ./docs collectionName my_docs modelname your_model_name targetChunkSize 1500
# Optional environment variable configuration
export QDRANT_URL=http://your - qdrant - host:6333
export QDRANT_API_KEY=your_api_key

# Document Retrieval
bun run query --run "Your search question"
# Advanced options
QDRANT_COLLECTION=my_collection \
NUM_RESULTS=10 \
EMBEDDING_MODEL=another_model \
bun run query --run "Advanced search question"

📚 Documentation

MCP Protocol Integration

By implementing the MCP protocol, this system can seamlessly connect with development environments that support the protocol, such as VS Code and IntelliJ IDEA. Users can directly trigger the document retrieval function during code editing to quickly obtain the required technical information.

Scalability Design

• Multimodal Support: There are plans to add the ability to process content types such as images and audio in the future.
• Distributed Architecture: A microservice - based deployment solution using Kubernetes is being planned.
• Incremental Indexing: Real - time file change detection and automatic index update functions are under development.

🔧 Technical Details

Document Indexer

• File Parsing: Supports Markdown format and automatically identifies and extracts structured content such as code blocks and paragraphs.
• Text Chunking: Performs intelligent chunking based on the target word count to ensure content integrity and retrieval efficiency.
• Vector Generation: Uses the Transformers.js library to convert text into high - dimensional vector representations.

Retrieval Engine

• Similarity Calculation: Uses cosine similarity for vector comparison and returns the most relevant document fragments.
• Result Sorting: Dynamically sorts retrieval results based on the matching degree.
• Context Expansion: Improves the relevance of retrieval results based on the information of adjacent chunks.

📄 License

Not provided in the original document.

Maintenance and Optimization

Performance Tuning

It is recommended to adjust the vector dimension and chunking strategy according to the actual hardware configuration. You can perform performance tests using the following command:

npm run benchmark

Log Management

The system provides detailed logging functions, which are convenient for troubleshooting and monitoring the running status.

Project Contribution

Welcome community developers to participate in the project development! Please submit code modifications or report issues in the following ways:

• GitHub Repository: [your - github - repository]
• Mailing List: [your - email - list]