Player not loading? Watch on YouTube
The original video is in French. This tutorial explains local document retrieval with QMD and LEANN, then connects both tools to Claude Code through MCP. The speaker uses an export of more than 18,000 blog articles to demonstrate searches and answers based on retrieved documents.
QMD focuses on Markdown. The walkthrough covers installation with npm, creating a named collection from a folder, indexing files and generating vectors. It compares the search command with query, which combines lexical and vector search, and shows how get retrieves a complete document. The speaker reports an error at the end of vector generation but says the process completed and searches worked in his setup.
The Claude Code demonstration uses QMD to answer questions about the speaker's articles. This separates local retrieval from the coding assistant that composes the answer; retrieved document content is passed to the external Claude service, so this demonstrated answer workflow is not entirely offline.
LEANN receives a shorter installation walkthrough using uv and a Python virtual environment, followed by command-line search and MCP configuration. The speaker describes it as more complex to set up and able to accept PDF and text alongside Markdown. He also reports slow searches on a separate, much larger dataset. Preserving URLs, dates and other context in exported documents helps the assistant identify sources, although a request for the latest article returns a result whose date the speaker questions.
