ocr

A minimalist OCR library for Rust with CLI and MCP server interfaces. Recognition is implemented in Rust (bitmap font templates, segmentation, and matching)—no Tesseract or other external OCR engine.

The MCP server is built on rmcp.

Features

• No native OCR dependencies - Pure Rust pipeline (binarization, line/character segmentation, template matching)
• Library - OcrEngine API with structured results (text, words, confidence, bounding boxes)
• CLI - Text and JSON output; optional preprocessing
• MCP server - Model Context Protocol server for assistant integration (stdio transport)
• Preprocessing - Optional grayscale + threshold preprocessing
• English-focused - Bitmap glyph set suited to Latin letters, digits, and common punctuation

Installation

cargo build --release

Binaries (from this crate):

Usage

Library

Crate name: ocr (import as ocr, not ocr_rs).

use ocr::OcrEngine;
use std::path::Path;

let engine = OcrEngine::new()
    .language("eng")
    .preprocessing(true);

let result = engine.recognize_file(Path::new("document.png"))?;
println!("{}", result.text);
println!("Confidence: {:.1}%", result.confidence);

CLI

# Plain text to stdout
ocr document.png

# JSON with word-level detail
ocr document.png -f json

# Enable preprocessing
ocr document.png --preprocess

MCP server

Add to your MCP client configuration (e.g. Claude Desktop):

{
  "mcpServers": {
    "ocr": {
      "command": "ocr-mcp",
      "args": []
    }
  }
}

Exposed tools:

Architecture

See ARCHITECTURE.md for design details and SPEC.md for the API specification.

License

Apache-2.0