Convert any webpage to clean Markdown
The webpage to Markdown converter turns the article on a web page into tidy Markdown you can paste into Obsidian, Notion, Logseq, a GitHub README, a static-site generator or a prompt for an AI assistant. Paste a link and you get the headline, paragraphs, sub-headings, lists, quotes, code, tables and pictures — and none of the menus, cookie banners, adverts, share buttons or footers around them.
Every link and picture address is made absolute, so the Markdown still works when it lives somewhere else. Choose inline links or reference-style links collected at the end, keep pictures as Markdown images or as their descriptions only, and add a YAML front-matter block with the title, source address and date for your notes app.
A fast built-in detector finds the article. When a page fools it, the optional Clean with AI (TypeSafe) button asks TypeSafe’s Jev model to score each block, and you can still tick or untick anything by hand.
How to convert a URL to Markdown
- Paste the page address and press Convert to Markdown (or Enter). The Markdown appears below a few seconds later.
- Choose how Links and Pictures are written, and whether to add Front matter. The Markdown is rebuilt instantly.
- If something is missing or extra, press Clean with AI (TypeSafe), or open Page blocks and tick or untick paragraphs yourself.
- Edit the text in the box if you like, then press Copy Markdown or Download .md.
Webpage to Markdown features
Reader-mode extraction
The page is split into blocks and a text-density detector — the approach browsers use for reader view — keeps the article and drops navigation, sidebars and footers.
Optional AI clean-up
TypeSafe’s Jev model scores each block from “site furniture” to “main article content”; blocks scoring 50% or more stay, and a picture stays when the text on both sides of it does.
Faithful Markdown
Bold, italics, inline code, strikethrough, nested and numbered lists, quotes, fenced code with its language, and GitHub-flavoured tables, written with the turndown library (MIT licence).
Absolute links and pictures
Relative addresses are resolved against the final page address, and jump links such as #notes point back to the original page.
Reference-style links
Keep paragraphs readable with [text][1] links and a numbered list of addresses at the end.
Front matter for notes apps
An optional YAML block with title, source, author and publication date (when the page states them) and the date you saved it.
What the Markdown looks like
A typical news article comes out like this, ready for any CommonMark or GitHub-flavoured Markdown reader:
| On the page | In the Markdown |
|---|---|
| Headline and sub-headings | # Headline, ## Section, ### Sub-section |
| Paragraph with a link | … the [Gulf Stream](https://…/gulf-stream/) … |
| Bulleted and numbered lists | - item and 1. step, nested lists indented |
| Quote | > quoted text |
| Table | | Current | Typical flow | with a | --- | header line |
| Picture with a caption |  followed by the caption in italics |
Built-in detection or AI?
The built-in detector scores the boxes on the page by how much running text they hold: long paragraphs and commas count for, a high share of link text counts against. It picks the best box and drops anything inside navigation, sidebars, footers and forms, or in boxes named like “share”, “related”, “newsletter” or “promo”. It is instant and sends nothing anywhere.
Unusual layouts can fool it, as can sign-up pitches placed inside the article. The AI pass reads the blocks themselves: in Galaxy of Tools’ evaluation on a news page and a recipe page with 56 blocks between them, it kept all 29 article blocks and dropped all 27 furniture blocks. On very long pages it judges the 400 blocks around the article and the rest keep the built-in decision. Its score appears next to each block in the list.
Privacy and limits
- The address goes to Galaxy of Tools’ page fetcher, which downloads the public page without cookies and refuses private network addresses. Pictures are only linked in the Markdown, never downloaded.
- With the AI button, the text of the blocks (up to 400 blocks, each cut to 400 characters), their kind and link share, the page title and address go through Galaxy of Tools’ server to TypeSafe. Nothing is stored.
- Pages drawn by JavaScript, or behind a login, paywall or cookie wall, can’t be read. Save such a page in your browser and convert the file with HTML to Markdown.
- Need a formatted document instead? Webpage to PDF uses the same reader mode, and Markdown to PDF typesets your Markdown.
Webpage to Markdown FAQ
How do I convert a webpage to Markdown?
Paste the page address into Webpage to Markdown and press Convert to Markdown. The article appears as Markdown in the box; copy it or download it as an .md file. It’s free and needs no account.
Does it keep links and images?
Yes. Links become [text](address) — or reference-style [text][1] links if you prefer — and pictures become , all with full absolute addresses. You can also keep only the picture descriptions or leave pictures out.
Can I use it as a web clipper for Obsidian or Notion?
Yes. Tick Front matter to start the file with a YAML block holding the title, source address, author, publication date and the date you saved it, then download the .md file into your vault or paste it into a page.
What exactly does “Clean with AI” send?
Only the text of the page’s blocks (each cut to 400 characters), what kind of block each is and how much of it is link text, plus the page title and address. TypeSafe’s Jev model returns a score per block; nothing is stored.
Why is the Markdown missing part of the article?
The detector judged that part to be page furniture, usually because it is mostly links or sits in a box named like a sidebar or promo. Open Page blocks, switch to Dropped, tick it, and the Markdown updates.
Why does it say no readable text was found?
The page is built by JavaScript or needs a login. Scripts don’t run here and nothing is logged in. Open the page in your browser, save it with Ctrl+S once it has loaded, and convert the saved file with HTML to Markdown.