Your model is reading div soup. Feed it Markdown.
LLMFeeder strips nav, ads, and script noise from any page, then copies the rest as clean Markdown. One click, all in your browser.
10.4× fewer tokens on the median page we measured. See the measurements
Add to Chrome
Get for Firefox
Free · MIT · v2.2.0 · No account, no telemetry
Scroll to strip a pageStrip a page to Markdown
On the measured MDN page, estimated tokens fall from 100,238 for the raw page to 8,577 for the Markdown LLMFeeder produces.
- Clear popups and banners
- Keep the main article
- Drop styles and chrome
- Rewrite it as Markdown
- Copy it. Paste anywhere.
Why context quality beats context length
A bigger context window does not make a model read better. It just makes it easier to send junk.
Raw page markup carries class names, tracking attributes and script tags that mean nothing to a language model.
- Markup is noise the model still has to read
- Structure is signal worth preserving
| Format | Tokens |
|---|---|
| Raw HTML | 100,238 |
| Markdown | 8,577 |
Measured with this repository's own readability.js and
turndown.js on
MDN's animation-timeline page
(100,238 to 8,577 est. tokens). Across five pages: median
10.4× reduction (4.5× to 15.1×). Estimates use the
extension's own ~4 bytes/token heuristic, not a billed tokenizer.
Same page. 7% of a 128K window, not 78%.
Tokens you skip on markup are free for the actual question. On a 128K context window, that measured page still leaves most of the room.
Built for cleaner pastes.
Pull the right content, keep the structure, and check the size before you send it.
- Smart extraction Mozilla Readability pulls the article, not the sidebar and ads around it.
- Multi-tab batch Convert several tabs at once. Merge into one file, or download a ZIP.
- Token awareness Live token estimates with configurable limits, before you paste.
- Tables stay intact HTML tables become Markdown tables, empty cells included.
- Lazy chat scroll Optional auto-scroll so long Gemini, ChatGPT, and Claude threads are not cut off.
- Keyboard first Copy, download or zip without opening the popup: AltShiftM / D / Z
- Links on demand Strip URLs to save tokens, or keep them when you need sources.
- Metadata you control Template a citation block with title, URL, author, date, and more.
The page never leaves your browser.
Extraction, Markdown conversion, and clipboard writes all run locally. No backend, no account, no upload.
- Nothing is sent over the network during conversion
- Only the permissions it needs to work
- MIT licensed, and you can read every line
Stop pasting markup.
Install it for your browser, then star the repo so the next person stuck in
<div> soup can find it too.