A browser extension that lets Anthropic's Claude see and interact

Android Browser Link

A browser extension that lets Anthropic’s Claude see and (and eventually) interact and edit web pages and the browser itself. It uses Claude.ai’s text field directly in a background tab, so no API key is required. Works with free accounts!

How it works

After installing (see below) visit any page and click the browser button

Claude.ai/chat will open up and the icon will turn green

Disconnected Connected

Press Scan to load the HTML into the page (this will be automatic)

Ask anything about the page

Features coming soon

  • Automatically scan HTML/Content based on query (right now you gotta do it manually)
  • Click on links and navigate pages on its own
  • Fill form fields and interact with page (ex: like and retweet on Twitter)
  • Know how to Google search on its own
  • See About this project below for more or follow the project in my technical diary: https://zactolle.super.site/projects/chatbot-browser-link

Coming soon! It’s now working but just needs some more documentation

Manual installing

  1. Download this project
  2. Sideload the extension to Edge, Chrome, Brave, or other Chromium browsers

1. First get the files

Method A: Download as .zip

(coming soon)

Method B: Download as git

Run the following from your terminal:

# Download
git clone https://github.com/zactolle/android-browser-link

# (Optional) Only if you want to develop the extension,
# it's already built so you can skip this
cd clause-browser-link
npm install
npm start
npm run build

2. Sideload the extension

Once you have the files, add the /dist folder to your browser. The instructions are similar for chromium browsers like Microsoft Edge and Google Chrome

  1. Enable Developer mode
  2. Load the /dist folder as an “Unpacked extension”
  3. (Optional) Make sure it is enabled
  4. (Optional) Click reload if you get errors

Developing

  1. Requirements
    1. Node.js is required for starting the local development server with npm start
    2. git is optional and needed to pull the latest version with git pull
  2. Debugging
    1. This is all still experimental and it crashes in weird ways sometimes. To “clear the cache bugs” reload the extension in chrome://extensions

About this project

The Problem

  • Claude is a foundational Large Language Model with a 100K context window, but no API
    • 100k is roughly the average length of 1 novel, 5 short stories, or several npm packages and codebases
  • If given access, Claude could read entire pages and edit the code directly using prompts/words
  • Since there’s no API, the current process is painful:

graph TD
save["Right click save"]-->flatten["Flatten to one file"]-->upload["Upload to Claude"]-->process["Process"]-->saveToFile["Save response as an .html file"]-->test["Test in browser"]

  • Even with API doing anything useful would involve knowing how to code

The Solution

  • Browser extension that syncs Claude.ai’s input field in one tab with the HTML of a separate tab so that you can ask Claude questions and have it directly affect the page
  • The Browser extension updates the input field on Claude.ai directly so no need to use API and can be used with a free account
  • Claude can affect the page directly, enabling automation and “breakout” use cases that use native browser APIs

Use Cases

  • Automate any browser task by just describing it with words, no coding required
    • Fill and submit form fields
    • Answer captchas
    • Automate any real world task that can be done by you with just a browser using just words
    • Build “worker bots” that can automate paid tasks on Fiverr
    • Analyze and ask questions about users, lists of users, hashtags, trending topics, and more just by visiting the pages
    • Permanently update a websites design or functionality by describing it with words

Vision

  • Browser Copilot
  • Browser AutoGPT
  • AGI browser extension

Contributing

This project is still mostly in the research and pre-validation phase, but feel free to create a Pull Request or ask for a feature or anything else. I’ll share links to a Discord soon!

GitHub

View Github