mcp-qdrant-docs MCP Server
@kazuph
An MCP server that scrapes websites, indexes content into Qdrant, and provides a query tool.
Overview
What is mcp-qdrant-docs MCP Server?
A TypeScript-based MCP server that scrapes website content, indexes it into a Qdrant vector database, and provides a tool to answer natural language questions about the indexed content. Designed for developers who want to query documentation or web pages using an AI assistant.
How to use mcp-qdrant-docs MCP Server?
Install globally with npm i -g @kazuph/mcp-qdrant-docs or use npx in your MCP client configuration. Configure in Claude Desktop’s claude_desktop_config.json by setting command, args, and environment variables such as DOCS_URL, QDRANT_URL, and COLLECTION_NAME. Then invoke the tool (e.g., ask_reactrouter_docs) with a natural language query.
Key features of mcp-qdrant-docs MCP Server
- Scrapes websites and indexes content into Qdrant.
- Dynamically generates a tool per website (e.g.,
ask_reactrouter_docs). - Embeds queries and searches Qdrant for relevant chunks.
- Supports force reindexing and limits on pages scraped.
- Configurable via CLI options and environment variables.
Use cases of mcp-qdrant-docs MCP Server
- Ask questions about React Router documentation using natural language.
- Query Hono.dev documentation content directly.
- Interrogate Gradio.app LLMs documentation.
- General Q&A over any website’s documentation.
FAQ from mcp-qdrant-docs MCP Server
What does the server do with scraped content?
It processes, chunks, and embeds the content using a sentence transformer model, then stores embeddings and chunks in a Qdrant collection specific to the website.
How do I configure the server for a new documentation site?
Set the DOCS_URL environment variable (or --start-url CLI option) and optionally QDRANT_URL, COLLECTION_NAME, and EMBEDDING_MODEL. Add an entry to your MCP client configuration file.
What are the dependencies/runtime requirements?
Requires Node.js and npm. The server needs access to a running Qdrant instance (default http://localhost:6333) and uses sentence transformer models from Hugging Face.
Where does the data live?
Scraped content and embeddings are stored in a Qdrant collection whose name is based on COLLECTION_NAME and the sanitized hostname of the scraped site.
What is the transport/auth method?
The server communicates via MCP protocol over stdio. No authentication is mentioned in the README.
