ChromeBoost

Privacy Policy

Last updated 4 September 2026

The short version. ChromeBoost has no servers of its own and collects nothing about your browsing. The free extension and MCP server talk only to your own computer. The paid desktop app adds exactly three outside connections, and each one is named below: Anthropic, because the app runs on your own Claude subscription; Polar, which handles payment and licence keys; and GitHub, which hosts the update check. We never see your prompts, the pages you automate, or the files you produce.

The one thing to understand before you start is in section 4: page content you ask ChromeBoost to read is sent to the AI you are using, under that provider's terms and privacy policy, not ours. Treat it the way you would treat pasting a page into a chat window.

1. Who this covers

This policy is published by Beaverhand Inc. ("we"). It covers four things:

The extension and MCP server are open source. If any statement here about them does not match the code, the code is the truth and the difference is a bug worth reporting. The desktop app's source is not public, so for the desktop app this policy is the description you can hold us to.

Contents

  1. Who this covers
  2. The extension and MCP server
  3. The desktop app
  4. Where page content actually goes
  5. Extension permissions
  6. What is never done
  7. The website
  8. Retention and deletion
  9. Your rights
  10. Children
  11. Changes
  12. Contact

2. The extension and MCP server

Nothing is collected by us. Neither the extension nor the MCP server has a backend, a logging service, a crash reporter, or analytics. There is no network destination of ours that data could be collected at.

The extension opens a single outbound connection, and it is a local one:

DestinationWhat travels over it
ws://localhost:PORT The MCP server on your own machine. Commands from your AI agent travel in, and the results of those commands travel back. This connection never leaves your computer.

The extension keeps a small amount of state in chrome.storage.local. It stays in your browser profile and is never transmitted: the position, collapsed state and visibility of the on-page status panel; which local agent sessions are connected and on which ports; and a short random prefix used to tag elements during a page operation. No browsing history, bookmarks, form data or credentials are stored by the extension.

3. The desktop app

The desktop app wraps the same extension and MCP server in a window so you can give instructions in plain English without a terminal. It runs on your own Claude subscription and adds a paid licence. Here is everything it stores and every outside connection it makes.

3.1 What stays on your computer

WhatWhere
Your jobs: the instructions you typed, the assistant's replies, and a record of each browser stepThe app's data folder: ~/Library/Application Support/ChromeBoost on Mac, %APPDATA%\ChromeBoost on Windows
Settings, repeat schedules, your licence key and activation id, and the count of finished jobs this monthThe same data folder
A short diagnostic log of what the app did: its version, when Chrome linked, when jobs started and ended, update checks, and any errors. It never contains page content or your instructionsThe same data folder, logs/
Crash reports, if the app ever crashes. They stay on your computer and are never uploaded; they exist so you can send one to us if you choose toThe same data folder, Crashpad/
Claude's own sign-in state, kept by Anthropic's software in a folder the app reserves for it, and in your operating system's keychainThe same data folder, engine/, plus the system keychain
Files you attach to a job, and files a job produces for you ChromeBoost Files in your home folder
A copy of the browser extension, when you install it from the app rather than the Chrome Web StoreChromeBoost Extension in your home folder

None of this is uploaded to us. We have no way to read it.

3.2 The three outside connections

WhoWhenWhat is sent
Anthropic (Claude) When you sign in, and during every job Sign-in happens in your own browser on Anthropic's site. The one-time code you paste into the app is passed straight through to Anthropic's own command-line software; the app does not read, parse or store it. During a job, your instructions, the page content the assistant asks to read, and its replies travel between Anthropic's software on your computer and Anthropic's servers, under your own Claude account. See section 4.
Polar (payments and licences) When you buy, when you enter a licence key, and periodically afterwards Buying happens on Polar's checkout page in your browser; Polar is the merchant of record and handles your payment details, receipts and tax. We never see your card number. When you enter a licence key, the app sends Polar the key, our organisation id, and your computer's name (so you can tell your own machines apart in Polar's portal), and receives an activation id. Afterwards the app re-checks the key and activation id with Polar from time to time; if you are offline it keeps working for 14 days without a check.
GitHub (update check) At launch and about every six hours while the app is open A plain request for a small text file that names the newest version. GitHub sees your IP address and a standard browser-style request, as any web host does. No account, identifier or usage data is included. Downloads of new versions come from the same place.

That is the complete list. The app has no telemetry, no analytics, no crash reporting to us, and no account with us. Usage limits on the free tier are counted on your computer and never reported anywhere.

3.3 What we receive, and from whom

The only information we hold about desktop app customers comes from Polar: the email address you gave at checkout, what you bought, when, the licence keys issued to you, and the names of the computers you activated them on. We use it to run your subscription, answer support requests, and meet tax and accounting obligations. We do not receive anything about how you use the app.

4. Where page content actually goes

Read this part. ChromeBoost is the hands, not the brain. When you ask for a page to be read, a form filled or a button found, the relevant page content is handed to the AI assistant you are using. In the desktop app that is Claude, running on your own Anthropic account. With the plugin it may be Claude Code, Codex or Gemini CLI. Those tools send what they receive to their provider's servers in order to think about it.

So while ChromeBoost itself sends nothing to us, the practical result of using it is that pages you point it at can reach Anthropic, OpenAI or Google, governed by their privacy policies and your agreement with them, not this one.

Two things we do to keep this deliberate:

The desktop app also pauses and asks you before actions that publish, pay or sign in, unless you switch that off in Settings. That is a design choice for your benefit, not a promise about what any site will do with the result.

5. Extension permissions, and why each one exists

Chrome shows a blunt warning at install time. Here is the specific reason for each permission the extension requests:

PermissionWhy it is needed
debuggerThe core of the product. Clicks, drags and keystrokes are dispatched through the Chrome DevTools Protocol so they arrive as real, trusted input events. Ordinary synthetic events are silently ignored by many modern web apps.
<all_urls>You decide which sites to automate, so the extension cannot know the list in advance. It acts only on the tab you direct it to.
tabsListing, opening, switching and closing tabs on request.
activeTabReading and screenshotting the tab currently being worked on.
scriptingInjecting the content script that performs DOM work such as hit testing, form filling and shadow DOM traversal.
offscreenHolding the WebSocket connection open. An MV3 service worker is shut down when idle, which would drop the connection.
storageRemembering the status panel position and which local sessions are connected.
windowsFocusing and arranging browser windows during multi-tab work.
downloadsSaving a file when you explicitly ask for a download, and reading back files the assistant was told to fetch.

6. What is never done

7. The website

The site is static and hosted on Vercel. Three things about it differ from the software, and it would be misleading not to say so:

8. Retention and deletion

9. Your rights

Depending on where you live, you may have rights to access, correct, delete or export personal data about you, to object to or restrict its processing, and to complain to a data protection authority. Because we hold so little, exercising them is simple:

Our lawful basis for the little processing we do is performance of the contract with you (running your subscription and support) and our legitimate interest in keeping the website working and understanding aggregate traffic. We do not make automated decisions about you. Polar and Vercel may process data outside your country under their own safeguards.

10. Children

ChromeBoost is not directed at children. The extension and MCP server are developer tools for people 13 and over. The desktop app is a paid product for people 18 and over, or the age of majority where you live if that is higher. We do not knowingly collect information from children; if you believe a child has given us any, contact us and we will delete it.

11. Changes to this policy

If this policy changes in a way that affects how data is handled, the date at the top will change and the revision will be visible in the public commit history of this file. If a change materially reduces your privacy, the desktop app will also tell you in its update notice before the new version applies.

12. Contact

Privacy questions, corrections, or requests about your data: me@lordamdal.com. For the open source parts you can also open an issue at github.com/lordamdal/chromeboost/issues. For suspected security issues, see SECURITY.md.