elliotroe/lit-bucket

Lit Bucket turns a Zotero research library into something an AI assistant can search. Zotero is reference management software researchers use to collect and organize academic papers, and this project watches the PDF attachments that Zotero syncs, converts each paper into Markdown text on the same machine using a tool called Docling, and builds a searchable index of that content. The index is organized by section heading, so a search can point directly at something like a results section or a subgroup analysis inside a specific paper, rather than just returning a whole document. Once the index is built, a separate part of the system called an MCP server lets an AI client, such as Claude, search across the whole library and pull the full text of any paper it finds relevant. Everything, meaning the file syncing, the PDF parsing, the search index, and the MCP server, runs together inside a single Docker container, so there is only one thing to deploy and run. Storage can be either Cloudflare R2, a cloud storage service, or a plain folder on the local machine, chosen with one configuration variable. Setting it up involves cloning the repository, copying an example environment file and filling in details like a WebDAV password and, if using R2, cloud storage credentials, then starting everything with Docker Compose. The first run downloads about a gigabyte of machine learning models for parsing and embedding text, and the project recommends giving the container at least 4 gigabytes of memory since that parsing step is demanding. After it is running, Zotero is pointed at the container's WebDAV address so it can sync files in, and an MCP capable AI client is pointed at the search server's address so it can query the library. The project also offers a free hosted option for university affiliated or non-commercial users who would rather not run their own server. Lit Bucket is released under the Apache 2.0 license and requires Python 3.11 or newer.