Whole discussions, not just the page.
FileConcat Clipper is a browser side panel that turns what you are reading into Markdown and hands it to an open fileconcat.com tab, where it joins a bundle you can copy or download in one piece. It is for anyone who assembles reading material for an LLM and is tired of pasting pages in one at a time.
A thread comes out as a thread
Copying a page into Markdown is the easy half, and every bookmarklet does it. What is actually hard to get out of a browser is a discussion. Every author, score and level of nesting is kept, so a reply still reads as a reply to the thing above it rather than as one more paragraph.
On Hacker News the whole comment tree arrives in one request, however deep it runs. Measured on a 638-comment thread: 264,851 characters, nested eight levels down, none of which was on screen when the button was pressed.
A listing, one item at a time
A subreddit, a Hacker News front page, a YouTube channel or a search page lists what it has loaded, with a tap target on each row. Tap the ones you want and each is opened and read on its own, so a session of scrolling becomes a set of files rather than one flattened page of headlines.
Words that were never on the page
A watch page gives you the video's description and its full transcript, which is not text the page was showing to begin with. A channel's Playlists tab lists its playlists, and clipping one clips the videos it holds.
Peek reads the Markdown back inside the panel before you send it, so the two ways a read goes wrong are visible while they are still cheap to fix: the wrong page shows in the first block, and a transcript that stopped early shows in the last.
Then hand it over
Press Send and every finished clipping goes into your fileconcat.com tab at once, opening one if there is none. A send adds to the bundle rather than replacing it, and files with the same name replace each other, so a repository and the discussion about it can sit in one bundle, and re-sending a corrected clipping fixes it in place.
The thing you came for is not always text.
Four situations, starting with the one that is hardest to do any other way.
Watching is the slowest way to read
Three pages of one channel, and no click between them. The panel re-reads whatever the tab is showing, so the offer changes shape on its own.
What you came for is not on the page as text at all. The panel takes the video's description and its whole transcript, so a talk becomes something you can search, quote and ask a question of.
How databases actually store your data.md
One file, named after the video.
A discussion you were not in
Six hundred replies, thirty of them on screen. The whole tree comes out with its nesting intact, so you can ask what the disagreement actually was instead of scrolling for the comment everybody is quoting.
## Comments
**quietfox** - 2026-08-19
The rule everybody skips is the naming one.
**anmutig** - 2026-08-19
We tried enforcing it. Two of the four
rules were wrong inside a month.
**tern** - 2026-08-19
Which two?
Documentation, beside the code that uses it
Drop the repository in, clip the pages of the documentation that matter, and both land in one bundle. The answer comes back grounded in the pages you read rather than in whatever the model remembers about that library.
## Directory structure
|-- src/
| |-- db/page.rs
| |-- db/wal.rs
| `-- lib.rs
|-- Runtime in tokio - Rust.md
`-- Timeouts and cancellation - Tokio.mdOne subject, spread across the web
A newsletter, a reference page, a specification and a changelog are four page shapes and one file. What holds them together is your question, not the sites they came from.
<file path="Why we moved off Postgres.md">
source: "https://acme.dev/p/off-postgres"
</file>
<file path="Runtime in tokio - Rust.md">
source: "https://docs.rs/tokio/latest/tokio/runtime"
</file>
<file path="RFC 9114- HTTP-3.md">
source: "https://www.rfc-editor.org/rfc/rfc9114"
</file>
<file path="Changelog - 2026-08-14.md">
source: "https://api.acme.dev/changelog"
</file>Everywhere else, it reads the article.
If a page reads as an article, the panel offers to clip it. Mozilla's Readability picks the body and Turndown renders it, which covers Substack, Medium, documentation sites, news and blogs with no per-site code. Navigation, sidebars, cookie bars and footers are left behind.
- threads
Reddit and Hacker News, whole. Expanding a Reddit thread clicks its own "more replies" three rounds deep, and is off by default because it is slower.
- listings
Subreddits, front pages, channels and search pages, with a tap target on every row.
- video
Descriptions and full transcripts from a watch page, and the videos a playlist holds. Comments are an opt-in extra, off by default, because they cost up to 45% more tokens.
- articles
Any page that reads as one, on any site, with no per-site code behind it.
What it does not do.
Nothing is read from a page until you ask for it. There is no background crawling, no page is touched because you happened to visit it, and no clipping is sent anywhere except the fileconcat.com tab you are looking at.
There is no account, no sign-in, and no server of ours in the path. Clippings are held in your browser's own storage between the clip and the send, and the sent list is a receipt rather than a library: rows leave after seven days.
The privacy page covers the extension alongside the site.
Put it next to what you are reading.
Free, open source, and it works with the tool you already use. Clippings land in the same bundle as the files you drop in yourself.