second-brain MCP

Built for anyone drowning in saved papers and scattered notes. Saves any arXiv link or PDF, auto-upgrades abstracts to full HTML, OCRs every figure with Claude Vision, then indexes everything in DuckDB with semantic embeddings. Search across figure panels by gene name or p-value, not just document titles. Auto-wikilinks new notes to related content already in your vault. Implements Ebbinghaus scoring to compress stale notes into PNG snapshots at 60-90% token reduction while keeping frequently accessed material at full fidelity. Exposes get_context, search_figures, search_notes, save_article, and update_goals over stdio. Pure Markdown vault, so you can sync via iCloud or git and switch agents anytime. Runs nomic-embed-text locally via llama-server with BM25 fallback.