LLMFeeder: a one-click webpage-to-Markdown clipper for LLM context

The problem LLMFeeder solves, and who it is actually for

Copying a web page into a chat model by hand is a lossy operation. You select text, you drag in navigation and cookie banners, tables collapse into unreadable runs of words, and code samples lose their fences. The README states the extension exists to convert web page content to clean Markdown format and copy it to clipboard with a single click, and that framing is accurate about the scope: this is a clipboard tool, not a scraping framework.

The intended user is someone who already works in a browser tab and wants the page in a prompt. The README positions the output as context for ChatGPT, Claude and Gemini, and the store listing was renamed to Web to Markdown for AI with a description targeting those searches. If your job is to hand a model a documentation page, a blog post or a long chat transcript, the extension removes the manual cleanup step.

It is not aimed at people building ingestion pipelines. There is no documented API, no server component and no batch entry point in the README. The value here is the one click, not the throughput.

How extraction and Markdown conversion work

The README names the Readability algorithm as the extraction step, which is the same approach reader modes use: score the DOM, pick the main content container, discard chrome. That choice explains both the strength and the weakness of the output. On an article page the result is close to what you would have selected by hand; on an application UI it depends entirely on whether the content sits in something Readability recognises as the main node.

Conversion happens in the page context, and the README is explicit that LLMFeeder operates as a fully client-side extension with zero backend dependencies, with all operations performed within the browser sandbox. It also states there is no analytics or tracking and that the extension requests only activeTab, clipboardWrite, storage and scripting. Those four permissions are consistent with the claim: none of them grant network access to a remote service, and the whole codebase is open source for inspection.

Iframes get special handling. The v2.2.1 release notes say same-origin and srcdoc frames are read through the DOM, while cross-origin frames keep a link or warning instead of page-to-page HTML messaging. That is a deliberate retreat from a fragile technique, and it means embedded cross-origin content will appear as a placeholder rather than as text.

For code, v2.2.0 added conversion of generic pre elements into fenced Markdown code blocks with language detection. The README lists this as better code block support rather than perfect, and language detection from markup is guesswork in most cases.

Installing LLMFeeder in Chrome or Firefox and converting a first page

For normal use the README recommends the browser extension stores. Chrome users install from the Chrome Web Store and Firefox users from the Firefox Add-ons listing; both links are in the README badges. Nothing else is required, because the extension has no backend to configure.

If you want to run the source instead, the repository ships a Makefile that drives a build script. The targets are chrome, firefox and source, and the versioned variants require a version parameter. The Makefile itself shows how the version argument is consumed:

make chrome
make firefox
make source
make versioned-chrome version=2.2.1
make clean

After installing, open any article page and use the keyboard shortcut the README documents for a quick convert, Alt+Shift+M. The Markdown lands on your clipboard, and the README also documents Alt+Shift+D to download the result as an .md file and Alt+Shift+Z to download multiple tabs as a ZIP.

Two settings are worth changing before your first serious use. The Include Links toggle strips URLs from links, which the README suggests for reducing token usage. The token counter shows a real-time estimate against GPT-4 and Claude tokenizers with configurable context limits from 4K to 32K, so you can see whether the page will fit before you paste it. Treat the number as an estimate, which is how the README describes it.

Where LLMFeeder breaks down

The most honest limitation is written into the release notes. v2.2.0 added lazy-loading auto-scroll as an opt-in setting to scroll chat and AI pages before extraction so long conversations are not truncated. That setting exists because the default behaviour truncates them: content below the fold is never rendered, so it is never extracted. If you convert a long Gemini or ChatGPT thread without enabling it, expect an incomplete result.

Cross-origin iframes are the second gap. The v2.2.1 notes describe keeping a link or warning for those frames rather than extracting their contents. Embedded videos, third-party comment widgets and payment forms will not appear as text.

Readability's main-content heuristic is the third. On a page whose value is in a sidebar, a comparison table or a comment thread, the extractor may discard exactly what you wanted. The v2.2.0 release added an alternative copy mode, a split copy button with a one-click override for Full Page versus Main Content, which is the project's answer to this. It is a per-use override, not a smarter extractor.

Finally, the tool is manual by construction. There is no documented way to run a conversion from a shell, a CI job or a scheduled task. If you need to convert a hundred URLs, this is the wrong tool and the README does not pretend otherwise.

How LLMFeeder differs from MarkDownload and similar clippers

The related searches around this project are mostly other clippers: MarkDownload, MarkSnip, Markdowner, Web2MD. The meaningful difference is where the output goes and what the tool assumes about your destination.

MarkDownload is known for saving into a notes vault, and its related searches include MarkDownload Obsidian. That is a file-first design: the Markdown is an artefact you keep, and the workflow assumes an editor or knowledge base at the other end. LLMFeeder is clipboard-first. The README's one-click claim is about getting text into a prompt, and the features that follow from that assumption are the token counter, the context limit setting and the Include Links toggle for trimming tokens. Those three make little sense in a note-taking tool and a lot of sense when the destination has a context window.

The multi-tab features point the same way. v2.1.0 added Copy All, Download Merged and Download as ZIP across selected tabs, which is about assembling several sources into one prompt rather than filing pages one by one.

If your destination is a vault, a clipper built around files will fit better. If your destination is a chat box, the token counter and the link toggle are the parts you would otherwise rebuild yourself.

Maintenance, licence and what upgrading costs

The repository is not archived, and the last push was on 2026-09-05. Releases have been frequent: v2.1.0 in February 2026, v2.2.0 in July 2026 and v2.2.1 in August 2026. The v2.2.1 notes are narrow, covering iframe extraction and safer iframe placeholders, which suggests the project is in a phase of tightening existing behaviour rather than adding new surface area.

The licence is MIT, stated in the README badge and present as a LICENSE file at the repository root. MIT is permissive, so redistribution and modification are allowed subject to its terms. That is a factual note about the licence identifier, not legal advice; read the LICENSE file if you plan to ship a fork.

Upgrade cost is low for store users, since Chrome and Firefox handle the update. For anyone building from source, the Makefile targets and the build script are the interface, and the versioned targets require an explicit version argument, so a release build is a deliberate command rather than a side effect. The package.json declares only Jest and jest-environment-jsdom as dev dependencies and a single test script, so the development footprint is small. There is no runtime dependency tree to audit, which follows from the extension being client-side only.

Editorial conclusion

Adopt LLMFeeder if your workflow is manual: you read a page in Chrome or Firefox and want it pasted into ChatGPT, Claude or Gemini as Markdown without retyping or fighting a reader view. Skip it if you need unattended conversion, a scriptable CLI or a server-side pipeline, because the project is a browser extension with no headless entry point documented in the README. Before relying on it, verify two things yourself: that the auto-scroll option actually recovers the full text on the specific chat UI you use, since the README frames it as an opt-in workaround for truncated conversations, and that the token counter's estimate matches the model you are feeding, since the README describes it as an estimation with configurable context limits rather than an exact count.

Frequently asked questions

Official sources

1. jatinkrmalik/LLMFeeder on GitHub
2. License: MIT
3. Project website
4. README
5. Releases