PdfCraft

Evidence key: [CLI] = pdfcraft-cli 0.13.0 help and tools; [MCP] = live tools/list; [src]; [unverified].

Purpose and reference product

PDF viewer/editor (Adobe Acrobat Pro clone; parity target Acrobat DC 26.002). Repo storytold/pdfcraft, MIT OR Apache-2.0. Git HEAD dc41044ba78f6cef2ff4bdb14d19272f54ad6aea (2026-10-15). Workspace 0.8.0 vs installed CLI 0.13.0 (version skew, README.md). The artcraft-in-a-box web bundle ships this repo as "PrintCraft" v0.2.1, i.e. the product has been renamed or double-branded; mcruft must not assume the name.

Native file format

PDF itself (no separate container). pdfcraft-cos is a lazy, tolerant object layer with copy-on-write edits and incremental or full save; pdfcraft-model gives typed page views. Passwords supported (--password); encryption RC4/AES rev 2-6.

Readable / writable formats

• Read: PDF (incl. encrypted, XFA forms). doc_create / doc_create_multiple turn PNG, JPEG, JPEG 2000, TIFF, GIF, BMP and .txt into PDF [MCP descriptions]. Office import is absent [src: gaps.md gap 8].
• Write: PDF; PDF/A-2b and 3b conversion (pdfa_convert); pages as PNG/JPEG/TIFF (doc_export_images, page_render; CLI render also .pam); text (doc_export_text); Word .docx, PowerPoint .pptx (one slide per page), HTML, RTF via doc_export_office ("Office export weak" per gaps.md); XFDF/FDF/XML/CSV form and comment data.
• No SVG export. No font-outline switch: this is an editor of existing PDFs, so fonts are whatever the file embeds; new text goes through text_font_import (validates embeddable licences).

CLI surface [CLI]

Subcommands: info, render, text, edit, combine, extract, split, check, audit, batch, tools, run, mcp, ui. - edit <in> --out <out> [--rotate 1,3:90] [--delete 2,4] [--move 5:1] [--insert-blank 1] [--title ..] [--author ..] [--full]: options run in order; page numbers 1-based and refer to the document after earlier options. - extract <in> --pages 1,3,5-8 --out out.pdf, split <in> (--every N | --before 3,7) [--out-dir], combine a.pdf b.pdf --out c.pdf. - batch <files|dirs> --steps convert,ocr,optimize --out-dir DIR [--recursive] [--ppi] [--quality] [--dpi] [--language]. - run <tool> [key=value ...] [--root DIR] [--out image.png] or run --script steps.json [--report] where steps are [{"tool":"doc_open","args":{...}}, ...]. Tool names are the MCP tool names. Errors: pdfcraft-cli: <tool>: unknown argument "blank" (expected: dpi, from, height, ...), exit 1 (verified). Tools are stateless per process unless a script keeps the doc id; I did not test multi-step state. - tools prints all tools with JSON Schemas (175 entries; the same data is tools/list). - mcp [--root DIR] [--compact] [--automation-read-root ..] [--automation-write-root ..]: --compact lists a core set plus tool_search/tool_call.

Command / tool catalogue [CLI tools, MCP]

175 tools. Document: doc_open {path,password}, doc_create {pages,width,height | images | text}, doc_info, doc_save, doc_close, doc_set_info, doc_combine, doc_split, doc_batch, doc_inspect. Pages: page_rotate, page_delete, page_move, page_reverse, page_insert_blank, page_insert_file, page_extract, page_duplicate, page_replace, page_set_box, page_render, page_add_text, page_add_image, page_images, image_edit, image_save. Content: content_list/update/delete, text_extract, text_find, text_edit.* [exact id unverified], text_fonts, text_font_import. Plus bookmarks, forms (AcroForm/XFA), annotations, redaction, signatures, accessibility (check, tagging), OCR, optimizer, compare, measure, print, JavaScript, headers/footers/watermarks. Many tools map to a command id (page.rotate, file.save, page.insert, ...) usable by command_run.

MCP surface [MCP]

pdfcraft-cli mcp (0.13.0, 175 tools listed non-compact). Open documents are MCP resources: pdfcraft://doc/{doc}/info, /text, /page/{page}/text, /page/{page}/image. Edits are in memory and undoable (edit_undo) until doc_save. Page numbers 1-based. Source: pdfcraft-automation (headless tool set + MCP).

Rust crates (33 in crates/)

crate	role	bindable as library?	notable deps
cos, model, content, filters, crypt	object layer, page tree, content streams, stream filters, encryption	yes	flate2, aes
organize, create, edit	page ops, create from images/text, headers/footers/watermarks	yes	-
render	page to pixels	yes (renderer is borrowed: hayro)	hayro, lopdf (patched copies in vendor/)
export	PDF to HTML/docx/RTF	yes	flate2
fonts, forms, annot, xfdf, xfa, redact, optimize, preflight, measure, compare, a11y, audit, print	features	yes (pick per need)	print needs craft-print
ocr	text recognition	yes but needs model files	rten, ocrs
js, sign	Acrobat JS (boa), digital signatures	yes but heavy and risky	boa_engine, rsa, p256
engine	facade over everything	yes, very heavy	image, icu
automation	tool set + MCP	no	tempfile
platform, ui-egui, oracle, testkit	OS services, GUI, test	no	egui

Cargo.toml has exclude = ["vendor"]: patched hayro/lopdf live in vendor/ and a consumer must re-declare them in its own [patch.crates-io] (pydoccraft does).

Operations relevant to mcruft

operation	route
page add/remove/reorder	page_insert_blank, page_insert_file, page_delete, page_move, page_duplicate, page_reverse; CLI edit
embedded vs outlined fonts	n/a (existing PDF); no outline switch
export PNG / SVG	PNG/JPEG/TIFF yes (page_render, doc_export_images); SVG none
import PDF	native; merge via doc_combine / page_insert_file
import images	doc_create images:[...], page_add_image {rect}
text edit	page_add_text, content_update; editing existing text is flagged fragile (gap 3)
single-page PDF	page_extract / CLI extract --pages N

Known gaps vs Acrobat Pro [src: docs/gaps.md]

Renderer borrowed and fidelity unmeasured; editing existing text fragile on real files; OCR only unaccented Latin; Office export weak, Office import absent; PDF/A partial, no PDF/X or PDF/UA; signatures lack PKCS#11, timestamp servers, revocation; accessibility checker without full tagging tools.