SnapMD

Clip any webpage as clean, LLM-ready Markdown from your terminal.

_{Webpages · YouTube · GitHub · Reddit · Hacker News · StackOverflow}

Install  ·  CLI  ·  MCP  ·  API  ·  Sources  ·  Contributing

npx snapmd https://en.wikipedia.org/wiki/Markdown -o article.md

SnapMD fetches any webpage and converts it to clean, structured Markdown — with YAML frontmatter, smart content extraction, and zero config. Built for piping into LLMs, pasting into Obsidian/Notion/Logseq, or building knowledge bases.

Why SnapMD?

Every day you copy from the web into your notes or AI tools. Every time: broken formatting, lost links, garbage HTML. Existing tools are bloated or abandoned.

SnapMD is different:

  Zero bloat — Hand-written recursive DOM parser. No Turndown, no Readability.js. Single dependency.
  LLM-ready — YAML frontmatter (title, author, date, word count, reading time). Pipe straight to any LLM.
  Smart extraction — Built-in readability engine strips nav, ads, sidebars. Or target with --selector.
  6 source types — Webpages, YouTube (transcripts), GitHub repos, Reddit posts, Hacker News, StackOverflow.
  MCP server — Use as an AI tool in Claude, Cursor, Windsurf, ChatGPT, and any MCP-compatible client.
  Batch mode — Clip multiple URLs in one command.
  Universal — CLI + API + stdin + MCP for every workflow.

Install

# Try instantly (no install)
npx snapmd https://example.com

# Install globally
npm install -g snapmd

# Add to your project
npm install snapmd

Requires Node.js 18+

CLI Usage

# Clip a webpage to stdout
snapmd https://example.com/blog/post

# Save to file
snapmd https://example.com -o article.md

# Target a specific element
snapmd https://example.com -s ".article-body"

# Batch clip multiple URLs
snapmd https://example.com https://github.com/user/repo

# YouTube video transcript
snapmd https://youtube.com/watch?v=dQw4w9WgXcQ

# GitHub README
snapmd https://github.com/suhashollakc/snapmd

# Reddit post with comments
snapmd https://reddit.com/r/programming/comments/abc123/title

# StackOverflow question + answers
snapmd https://stackoverflow.com/questions/12345/title

# Hacker News thread
snapmd https://news.ycombinator.com/item?id=12345

# JSON output (markdown + metadata + stats)
snapmd https://example.com --json

# Custom HTTP header (e.g. auth)
snapmd https://api.example.com -H "Authorization: Bearer token"

# Pipe to an LLM
snapmd https://example.com -F | llm "summarize this"

# Read HTML from stdin
curl -s https://example.com | snapmd --stdin

All Options

MCP Server (AI Tool Use)

SnapMD includes an MCP (Model Context Protocol) server, so AI assistants like Claude, Cursor, and Windsurf can clip webpages as a tool.

Setup

Add to your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "snapmd": {
      "command": "npx",
      "args": ["-y", "snapmd-mcp"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "snapmd": {
      "command": "snapmd-mcp"
    }
  }
}

Available Tools

Programmatic API

import { clip, clipHtml } from 'snapmd';

// Clip a URL
const result = await clip('https://example.com/article');
console.log(result.markdown);  // Clean Markdown with frontmatter
console.log(result.meta);      // { title, author, date, url, ... }
console.log(result.stats);     // { words: 1234, readingTime: '5 min read' }

// Clip with a CSS selector
const result2 = await clip('https://example.com', {
  selector: '.post-body',
  frontmatter: false,
});

// Clip with custom headers
const result3 = await clip('https://api.example.com/page', {
  headers: { 'Authorization': 'Bearer token' },
});

// Convert raw HTML
const { markdown } = clipHtml('<h1>Hello</h1><p>World</p>', {
  frontmatter: true,
  baseUrl: 'https://example.com',
});

Output Format

---
title: "How Markdown Changed Writing"
source: "https://example.com/blog/markdown"
author: "Jane Smith"
date: 2025-12-01
word_count: 1847
reading_time: "8 min read"
clipped_at: 2026-03-03T10:30:00.000Z
tool: snapmd
---

# How Markdown Changed Writing

The first paragraph of the article...

What It Converts

Headings (h1-h6), bold, italic, strikethrough, links with URL resolution, images, ordered and unordered lists (nested), blockquotes, code blocks with language detection, GFM tables, checkboxes, details/summary, figures, and more.

Supported Sources

Architecture

snapmd/
  bin/
    cli.js                CLI entry point
    mcp.js                MCP server for AI tool use
  src/
    index.js              Core engine (fetch, route, convert)
    converter.js          HTML to Markdown (recursive DOM walker)
    readability.js        Content extraction (strips nav/ads/noise)
    frontmatter.js        YAML frontmatter generation
    parsers/
      youtube.js          YouTube metadata + transcript
      github.js           GitHub repo + README
      reddit.js           Reddit post + comments
      hackernews.js       HN thread + comment tree
      stackoverflow.js    SO question + answers

Only one runtime dependency: linkedom for server-side DOM parsing. Everything else is hand-written.

Contributing

Contributions welcome! Especially: new source parsers (Twitter/X, PDF), edge case fixes, and performance improvements.

git clone https://github.com/suhashollakc/snapmd.git
cd snapmd && npm install
npm test                                          # 106 tests
echo '<h1>Test</h1><p>Hello</p>' | node bin/cli.js --stdin

License

MIT © Suhas Holla

_{If SnapMD saves you time, consider giving it a ⭐ on GitHub}