harumi

Pure-Rust PDF — CJK font embedding, OCR text overlay, text extraction, HTML→PDF, page merge/split.

Why harumi?

Need	API
OCR invisible text layer on scanned PDFs	`add_invisible_text` · `add_invisible_text_runs`
Extract searchable text from existing PDFs	`extract_text_runs` · `extract_text_chunks` · `extract_as_markdown`
Watermark / stamp visible text	`add_text` · `add_text_with_rotation`
Merge or split PDFs	`merge_from` · `extract_pages`
Draw shapes (rect, line, ellipse, polygon, path)	`add_rect` · `add_line` · `add_ellipse` · `add_polygon`
Embed JPEG/PNG images with transparency	`add_image` · `add_image_with_opacity`
Convert HTML to PDF	`render_html_to_pdf` (`html` feature)
Use in WASM / Lambda / Edge (no C/C++ deps)	All APIs work cross-platform

Quick Start

Overlay invisible OCR text

use harumi::Document;

let mut doc = Document::from_file("scanned.pdf")?;
let font = doc.embed_font(include_bytes!("NotoSansCJK-Regular.ttf"))?;

// Invisible layer for search/copy
doc.page(1)?
    .add_invisible_text("日本語テキスト", font, [72.0, 700.0], 12.0)?;

doc.save("searchable.pdf")?;

Add visible watermark

// Red "CONFIDENTIAL" stamp centered on page
let (w, h) = doc.page(1)?.size()?;
doc.page(1)?.add_text(
    "CONFIDENTIAL",
    font,
    [w / 2.0 - 60.0, h / 2.0],
    24.0,
    [0.8, 0.0, 0.0], // RGB red
)?;

Extract searchable text

let runs = doc.extract_text_runs(1)?;
for run in runs {
    println!("Text: {} at ({}, {})", run.text, run.x, run.y);
}

Feature Flags

Flag	What it enables	Dependencies
(default)	Text overlay, font embedding, text boxes, text extraction	`lopdf`, `ttf-parser`, `getrandom`
`draw`	Shapes: rect, line, ellipse, polygon, polyline	none
`image`	JPEG/PNG embed, extract from scanned PDFs	`png` crate
`ocr`	Tesseract coordinate conversion helpers	none
`flow`	FlowDocument: auto-pagination, headers/footers	none
`html`	HTML→PDF conversion (h1–h6, p, table, ul/ol)	none

Supported Fonts

Format	Status
TrueType (`.ttf`)	✅ Full support — pure-Rust subsetting
TTC collections (`.ttc`, multiple faces)	✅ Full support — `embed_font_at(bytes, face_index)`
OpenType CFF (`.otf`)	⚠️ Accepted, no subsetting (full font embedded)

Recommended fonts (CJK):

NotoSansCJKjp-Regular.ttf (Japanese)
NotoSansCJKsc-Regular.ttf (Simplified Chinese)
NotoSansCJKtc-Regular.ttf (Traditional Chinese)
NotoSansCJKkr-Regular.ttf (Korean)

Installation

[dependencies]
harumi = "1"

For image or HTML features:

harumi = { version = "1", features = ["image", "html"] }

Use via MCP Server

Use harumi's PDF tools directly from Claude Code, Cursor, or Continue IDE:

# Build the MCP server (pure Rust, no runtime dependency)
cargo build -p harumi-mcp

# Configure in your IDE and use tools:
# - pdf_extract_text: Extract text with positions
# - pdf_add_invisible_text: Add searchable OCR layer
# - pdf_html_to_pdf: HTML to PDF conversion
# - pdf_merge: Merge PDFs
# - pdf_page_info: Get page count & dimensions

Register on: smithery.ai or mcp.so (coming soon)

Why choose harumi?

✅ Pure Rust — zero C/C++ dependencies; works in WASM, Lambda, cross-compile
✅ CJK-native — full support for Chinese, Japanese, Korean fonts
✅ Simple API — complex font subsetting happens automatically at save time
✅ Text extraction — decode CID fonts + standard fonts (Type1, TrueType, WinAnsi)
✅ Text replacement — rewrite Tj/TJ operators with automatic re-subsetting
✅ Rich features — draw shapes, embed images, page merge/split, HTML→PDF
✅ Well-tested — 100+ unit + integration + E2E tests

More Info

Full documentation — docs.rs/harumi
Live demo — Browser annotation editor (WASM)
Source code — github.com/kent-tokyo/harumi
License — MIT OR Apache-2.0

Roadmap

v1.x — Current stable
v2.0 — PDF/A compliance, true digital signature verification (RSA/ECDSA)
Future — AES-256 write encryption, RTL text (Arabic/Hebrew)

See the full README on GitHub for extensive examples, API reference, and internals explanation.

harumi 1.3.1