MCP Qdrant Docs

🚀 MCP-Quant Documentation

mcp-quant is an MCP server designed to scrape content from specified websites and index it into the Qdrant vector database. This server offers a tool that can answer questions about the documents using natural language queries.

🚀 Quick Start

💻 Usage Examples

Basic Usage

# Add a server instance specifically for Hono.dev documentation:
{
  "mcpServers": {
    // ... Other servers ...
    "hono_docs": {
      "command": "npx",
      "args": [
        "@kazuph/mcp-qdrant-docs"
      ],
      "env": {
        "DOCS_URL": "https://hono.dev/",
        "QDRANT_URL": "http://your-qdrant-instance:6333",  # Replace with your Qdrant address
        "COLLECTION_NAME": "hono_docs_hono_dev"  # Base name, actual will be "hono_docs_hono_dev"
      }
    }
    // ... Other servers ...
  }
}

After such configuration, the MCP client will provide a tool named ask_hono_docs (instead of ask_hono_dev_docs).

• Tool Name: ask_<doc name>_docs (e.g., ask_reactrouter_docs)
• Description: Questions about the documentation content of the specified site (DOCS_URL).
• Input: Natural language queries about the documentation content.

The server will automatically scrape the website (if the collection does not exist or --force-reindex is used) and index the content into the specified Qdrant collection. Then, it will answer queries using the index through the provided tool (ask_reactrouter_docs).

✨ Features

1. Automatic Scraping and Indexing

• Automatic Scraping at Startup: If the content of the target site is not indexed or --force-reindex is enabled, the server will automatically scrape content from the specified website.
• Processing and Embedding Text Fragments: Process the scraped content using the selected sentence encoder and embed it into the vector space.
• Storage in Qdrant Collection: Store the processed text fragments and their embeddings in the specified Qdrant collection.

2. Query Tools

• Dynamically Generated Tool Names: Tool names are automatically generated based on the hostname of DOCS_URL. For example:
  • If the input site is https://reactrouter.com/, the tool name is ask_reactrouter_docs.
• Natural Language Queries: Support natural language queries about the documentation content and perform vector similarity searches through Qdrant to return the most relevant text fragments.

3. Flexibility and Scalability

• Support for Multiple Models: You can choose different sentence encoders/embedding models (e.g., all-MiniLM-L6-v2, Xenova/all-MiniLM-L6-v2).
• Configurable Scraping Parameters: Support setting scraping depth, limiting the scope of crawling, etc., to ensure the integrity and standardization of content scraping.

📦 Installation

1. Installation

npm install -g @kazuph/mcp-qdrant-docs

2. Running with npx

npx @kazuph/mcp-qdrant-docs --start-url https://example.com/ --collection-name my_docs

3. Configuration File Example

Add a server instance in claude_desktop_config.json:

{
  "mcpServers": {
    // ... Other servers ...
    "react-router-docs": {
      "command": "npx",
      "args": [
        "@kazuph/mcp-qdrant-docs"
      ],
      "env": {
        "DOCS_URL": "https://reactrouter.com/",
        "QDRANT_URL": "http://your-qdrant-instance:6333",  # Replace with your Qdrant address
        "COLLECTION_NAME": "react-router-docs"  # Base name, actual will be "react-router-docs-reactrouter_com"
      }
    }
    // ... Other servers ...
  }
}

🔧 Technical Details

Command-Line Options

Required Parameters

• --start-url: Specify the starting URL for scraping.
• --collection-name: Specify the name of the Qdrant collection.

Optional Parameters

• --depth: Scraping depth, default is 1.
• --max-links: Maximum number of links to scrape, default is 50.
• --exclude-domains: Exclude specific domains, separated by commas.
• --force-reindex: Force reindexing of an existing collection.

Examples

1. Adding a React Router Documentation Server Instance

Add the following to claude_desktop_config.json:

{
  "mcpServers": {
    "react-router-docs": {
      "command": "npx",
      "args": [
        "@kazuph/mcp-qdrant-docs"
      ],
      "env": {
        "DOCS_URL": "https://reactrouter.com/",
        "QDRANT_URL": "http://localhost:6333",  # Assume Qdrant is running locally
        "COLLECTION_NAME": "react-router-docs"  # Collection name
      }
    }
  }
}

2. Starting the Server

npx @kazuph/mcp-qdrant-docs --start-url https://reactrouter.com/ --collection-name react-router-docs

📄 License

No license information provided in the original document.

⚠️ Important Note

• Ensure that the Qdrant service is running correctly and can be accessed via the specified address.
• If you need to scrape a large amount of content, it is recommended to limit the scraping depth and the number of links to avoid overloading the target website.
• Before using --force-reindex, make sure that the target collection does not exist or its data is no longer needed.

Summary

With the MCP server mcp-quant, you can easily index the content of specified sites into the Qdrant vector database and provide natural language query functionality. This tool supports flexible configuration and expansion, making it suitable for various document management and retrieval scenarios.