Site Metadata Extractor

Cleans and extracts a web(site) resource's metadata.

Metadata extraction fields currently supported:

Name	Data Type
author	array (jsonb)
canonical_url	string
copyright	string
date (publish date)	date
description	text
favicon	text
image (primary/og image)	text
jsonld (structured data)	object (jsonb)
keywords	array (jsonb)
lang	string
locale	string
origin	string
publisher	string
site_name	string
tags	array (jsonb)
title	string
type	string
truncated_text	text
status	string
videos	array (jsonb)
links	array (jsonb)

Install

NPM:

$ npm install site-metadata-extractor --save

Yarn:

$ yarn add site-metadata-extractor

Usage

Feed in a raw markup from a webpage to get extracted metadata fields.

Modern typed API

The typed API is additive. The default export remains available for existing callers, while new callers can pass already-fetched public HTML into:

import {
  extractFromHtml,
  extractMetadataOnly,
  extractLazy,
  type ExtractedResource,
} from "site-metadata-extractor";

const resource: ExtractedResource = extractFromHtml(html, {
  inputUrl: "https://example.com/requested-url",
  finalUrl: "https://example.com/final-url",
  lang: "en",
});

This package does not fetch network resources. Fetch HTML in the calling application, apply any SSRF/private-network protections there, then pass the HTML string into the extractor.

extractFromHtml(html, options) returns stable typed output for ingestion:

• URL fields: inputUrl, finalUrl, canonicalUrl, normalizedUrl, domain
• metadata: title, softTitle, description, author, publisher, siteName, lang, locale, publishedAt, modifiedAt
• assets: faviconCandidates, imageCandidates, primaryImage
• structured/raw data: jsonld, rawMeta
• content: links, videos, readableText, textStats
• extraction metadata: extraction.packageVersion, extraction.strategyVersion, extraction.warnings, extraction.confidence

extractMetadataOnly(html, options) returns the same shape but skips readable text, link, and video extraction.

extractLazy(html, options) uses instance-local caches and exposes: metadata(), readableText(), links(), videos(), and extract().

Exported output types include ExtractedResource, AssetCandidate, ExtractedLink, ExtractedVideo, TextStats, and ExtractionMetadata.

Migration note: the legacy default export still returns the historical field names such as canonicalLink, favicon, image, and text. New integrations should prefer extractFromHtml so candidate URLs are resolved against finalUrl/inputUrl, JSON-LD is always an array, oversized metadata is bounded, and malformed JSON-LD is reported in extraction.warnings instead of being logged.

From .html file:

import fs from "fs";
import siteMetadataExtractor from "site-metadata-extractor";

const getMetadataFromFile = (filename) => {
  const filepath = path.resolve(__dirname, `../data/${filename}.html`);
  const markup = fs.readFileSync(filepath).toString();
  // feel free to use localhost as the second parameter for testing
  const metadata = siteMetadataExtractor(markup, "YOUR_SITE_ORIGIN_HERE");
  return metadata;
};

getMetadataFromFile("example");

From a server request:

import axios from 'axios';
import siteMetadataExtractor from 'site-metadata-extractor';

const processSite = async (url) => {
  return axios.get(url, config = {})
    .then(res => {
      const { headers } = res;
      const contentType = headers['content-type'];
      if (contentType.includes('text/html')) {
        return {
          body: res.data,
          url
        };
      } else {
        return {};
      }
    })
    .catch(err => {
      console.log(err);
    });
};

processSite('https://www.cnbc.com/guide/personal-finance-101-the-complete-guide-to-managing-your-money/`)
	.then((data) => {
		...
    siteMetadataExtractor(data, "https://www.cnbc.com/guide/personal-finance-101-the-complete-guide-to-managing-your-money/", "en");
    ...
	});

Development

1. Run: git clone https://github.com/sc10ntech/site-metadata-extractor.git
2. Change into project directory and install deps: cd site-metadata-extractor && npm i

Creids & Disclaimer

site-metadata-extractor was inspired by, and tries to be the spiritual successor to node-unfluff