@jeseong77/markboost-seo

Scans Markdown files to extract metadata and auto-generate SEO information (titles, descriptions) for English content using local LLMs.

Author's Site / Example: Library applied website

What it does

@jeseong77/markboost-seo is a Node.js library that processes your Markdown content to:

• Extract titles and file paths.
• Generate URL-friendly slugs.
• Optionally, use local LLMs (@xenova/transformers) to generate SEO titles and meta descriptions for English content.
• Return structured data for easy use in Static Site Generators (SSGs) or other tools.

Installation

npm install @jeseong77/markboost-seo
# or
yarn add @jeseong77/markboost-seo

Quick Start

import { scanAndProcessNotes, ScanOptions } from '@jeseong77/markboost-seo';
import path from 'path';

async function main() {
  const options: ScanOptions = {
    contentPath: path.join(process.cwd(), 'my-markdown-vault'),
    generateSeo: true, // Enable LLM SEO for English content
    llmConfig: {
      minContentLengthForSeo: 50, // words
      // maxContentLengthForPrompt: 250, // words (default)
      // modelName: 'Xenova/LaMini-Flan-T5-783M' // Default model
    },
  };

  try {
    const result = await scanAndProcessNotes(options);
    console.log(`Processed ${result.allNotes.length} notes.`);
    result.allNotes.forEach(note => {
      if (note.seo) {
        // console.log(note.title, note.seo.description);
      }
    });
  } catch (error) {
    console.error("Error:", error);
  }
}

main();

Key Options (ScanOptions)

• contentPath: string: Required. Path to your Markdown directory.
• generateSeo?: boolean: Set to true to enable LLM-based SEO (default: false).
• llmConfig?: LLMConfigOptions: Configure LLM model, content length thresholds.
• ignorePatterns?: string[]: Glob patterns for files/folders to ignore.
• includeFileContent?: boolean: Include full Markdown content in results (default: false).
• forceScan?: boolean: Force re-scan, bypassing cache in production (default: false).

Output (ScanResult)

The scanAndProcessNotes function returns a ScanResult object containing:

• allNotes: ProcessedNode[]: Array of processed file data. Each ProcessedNode includes id, title, filePath, fullPathSlug, simpleSlug, and an optional seo object ({ title?: string, description?: string }).
• notesMapByFullPathSlug: Map<string, ProcessedNode>
• notesMapBySimpleSlug: Map<string, Set<string>>

Note on LLM SEO

• LLM processing is currently optimized for generating meta descriptions for English content.
• SEO titles from LLM often default to the original title.
• Performance on CPU can be slow for LLM tasks.

License

MIT - See LICENSE file.