QDrant Loader
A toolkit that collects content from Git, Confluence, JIRA, public documentation sites and local files, vectorizes it into a Qdrant database, and serves it back through an MCP search server.
The project is a monorepo of three packages. The loader connects to Git, Confluence Cloud or Data Center, JIRA Cloud or Data Center, public documentation sites and local files, converts formats such as PDF, Office documents, images, audio, EPUB and ZIP archives, chunks documents with hierarchical context, and tracks incremental changes so re-syncing does not require reprocessing everything. A core library provides a provider-agnostic interface over OpenAI, Azure OpenAI, Ollama or custom endpoints, along with configuration, rate limiting and structured logging shared by the other packages.
An MCP server then exposes search over the resulting Qdrant collections to AI development tools such as Cursor and Windsurf, supporting a recent MCP protocol revision over both stdio and HTTP transports, so an editor's AI assistant can search a team's documentation, tickets and code history as context.