Ninjadoc
Ninjadoc is an MCP server for PDF document extraction with visual provenance. When an LLM answers a question about a document, it gets the answer plus a screenshot of the highlighted region in the source PDF so citations are grounded in what's actually on the page.