# vastglint > Local macOS CLI, Finder Quick Actions, and stdio MCP server for indexing large multi-format book/PDF libraries and keeping files clean and organized on disk. Website: https://vastglint.com/ Languages: https://vastglint.com/ (English), https://vastglint.com/zh-Hans/ (简体中文), https://vastglint.com/zh-HK/ (繁體中文(香港)), https://vastglint.com/ja/ (日本語), https://vastglint.com/ru/ (Русский), https://vastglint.com/ar/ (العربية) Repository: https://github.com/slowsignallabs/vastglint Publisher: Slow Signal Labs (https://slowsignal.xyz/) License: MIT Requirements: macOS 14+, Xcode 16+ (Swift 6) ## Why use vastglint - Solves context-window bloat when working with 50-MB PDFs, scanned DjVus, and multi-book folders across Claude Code, Google Antigravity, ChatGPT Work, Codex, and stdio MCP clients. - Compiles books into structured `_llm_corpus/` bundles with `BOOK.md` line ranges (`BOOK.md#Lstart-Lend`), SQLite FTS5 + CJK 2-gram + radical normalization + Apple `NLEmbedding`, Apple Vision ANE OCR for scans and diagrams, recovered tables/code blocks, PDF annotations, and cross-book DOI/arXiv/RFC citations (~90% fewer tokens than raw PDF reads). - Cleans metadata and organizes libraries safely with dry-run previews, canonical naming, streaming XMP/OPF/EXTH/ComicInfo/xattr scrubbing, SHA-256 and content deduplication to macOS Trash (preserving Vol I/II sets and PDF+EPUB pairs), and reversible topic clustering (`--flatten`, `--undo`). ## Supported formats - Automatic (11 formats): PDF, EPUB, DjVu (`.djvu`/`.djv`), FB2 (`.fb2`/`.fb2.zip`), MOBI/AZW3/AZW/PRC, CBZ/CBR, DOCX, ODT, RTF, DOC. - Explicit / opt-in only: Markdown (`.md`) and TXT (`.txt`) stay untouched in folders and mixed `Cmd+A` Finder selections unless selected standalone or enabled with `--include-text`. ## Core CLI commands ```zsh git clone https://github.com/slowsignallabs/vastglint && cd vastglint && ./scripts/install_service.sh vastglint doctor vastglint inspect ~/Books/monograph.djvu vastglint build ~/Books --alias systems vastglint query @systems "leader lease" --figures --annotations vastglint pack @systems "clock assumptions" --mode contrast --format claude-xml --budget 12000 vastglint graph @systems --topic "raft" --compare vastglint link @systems --format claude-xml --budget 14000 vastglint clean ~/Books --organize --dry-run vastglint clean ~/Books --organize vastglint organize ~/Books --flatten vastglint organize --undo ~/Books vastglint watch ~/Books --alias systems --auto-clean --auto-organize vastglint mcp ``` ## MCP tools (`vastglint mcp`) - `corpus_status`: list registered/active corpora, book counts, tokens, figures, and annotations. - `corpus_toc`: table of contents with `BOOK.md#Lstart-Lend` line ranges and TF-IDF keywords. - `corpus_search`: hybrid FTS5 BM25 + CJK 2-grams/Radicals + trigram + `NLEmbedding` RRF search (`figures_only`, `annotations_only`). - `corpus_pack`: token-budgeted compiler (`focused`, `contrast`, `annotations`, `figures`; `claude-xml`, `markdown`, `json`). - `corpus_read`: read exact line range on `books//BOOK.md` (max 400 lines). - `corpus_inspect_figure`: caption, 2D OCR layout, `cited_on_pages`, asset path, lazy base64 PNG (`include_image=false` default). - `corpus_compare`: cross-book `arXiv`, `DOI`, `RFC`, and `(Author, Year)` citation graph and comparison excerpts. - `corpus_ingest`: incrementally add books or folders during an active session.