Chrome extension development company for tools that live inside the browser your team already uses.

Custom Chrome extensions on Manifest V3: AI side panels, content scripts that read or fill the pages you open, and companion extensions for your product. A focused extension costs $3,500 to $7,500 and takes about three weeks.

5 days
first build to v0.9 of our own extension
0
emails it sends without a person clicking Send
MV3
Manifest V3, the current Chrome platform
10+
years of React and TypeScript, the founder's record
A person using a laptop
An extension works where your team already works: inside the tab they have open.

A Chrome extension is software that lives inside someone else’s page. It sees what your team sees at the moment they see it, which makes it the right tool when the work happens in a CRM, a helpdesk, a job board or a marketplace that you do not control and cannot change. Built well, it removes the copying between tabs. Built badly, it asks for every permission, breaks the week a site changes, and fails store review.

Extensions come from the same team as our agentic AI, AI content and automation work. The founder, Babar Ali, has written React and TypeScript for more than ten years, and at EDGE he built the shared Web Worker and WebSocket layer that a set of micro-frontends used for live updates. An extension is the same kind of system: a popup, a content script, a background worker and often a page of its own, each one isolated and talking to the others only by message.

What we build

  • Sales and recruiting tools that find, rank and log leads in the pages your team already opens
  • AI side panels that summarise, answer or draft next to the page someone is working on
  • Content scripts that read a page or fill in a form, so data stops being copied by hand
  • Support tools that bring the customer record into the helpdesk tab
  • Internal tools for one team, installed by policy and never listed in the store
  • Companion extensions for a SaaS product, signed in with your existing accounts
  • Manifest V2 to V3 migrations for extensions Chrome has stopped running

Most of these start as one repeated task: a person copying a name, an email or an order number from one tab into another, many times a day. That task is the scope. Everything else is added only when the first version is in use.

What we do not build

  • Extensions that post, message or click on a site for someone without them pressing a button
  • Ad injection, affiliate link swapping or anything that quietly changes a page to earn from it
  • Tools that collect browsing history, or send page content anywhere the user did not agree to
  • Bots built to get around a site's security checks, rate limits or terms

Turning that work down costs us some projects. It is also why the extensions we do build pass review and keep the trust of the people who install them.

Our own extension, from first build to v0.9 in five days

All work

The best evidence we have for this service is a tool we built for ourselves: a lead and email automation that runs as a Chrome extension on Manifest V3. It went from a first build to version 0.9 in five days, between 30 October and 3 November 2026.

It reads only the pages the user opens, read-only, and only after they press Start. It scrolls at a human reading pace, keeps the items that carry an email address, and hands them to a leads page with lists, a board, stats and keyboard triage. AI ranks the leads against a criterion the user types and can sort them into lists by each list’s own rules. The AI calls go to an OpenAI-compatible API, so a local model works as well as a hosted one.

Sending happens outside the browser. A small Node.js helper runs on 127.0.0.1 and answers only the extension and its own pages. It sends from several Gmail accounts over SMTP, reads replies over IMAP, rotates between the accounts, keeps a daily cap on each, never writes to the same address twice, and runs a deliverability test that checks where a message landed and what SPF, DKIM and DMARC said about it. Nothing sends until a person clicks Send.

One lesson from that build is now a rule on every extension we make. A popup closes the moment the user clicks away, and a message it fires on the way out can simply be lost. So anything a popup starts is written to chrome.storage first and picked up by the background service worker, including run plans of several searches that it spaces out with chrome.alarms.

lead-run · a run plan of three searches
09:00:04  Start pressed · search 1 of 3 in this tab
09:00:31  page read at reading pace · items with an email kept
09:06:12  break over · search 2 starts (chrome.alarms)
09:19:40  leads ranked by AI against your criterion
09:19:41  drafts ready · waiting for a person to review
✓ nothing sends until someone clicks Send

An illustration of one run, not an export of real data. The order of the steps is how the tool works; the times are examples.

lead-run · today
step 1  leads read from the pages opened
step 2  ranked by AI against the criterion
step 3  drafts ready, waiting for Send
✓ 0 emails sent without a click
Lead and email automation that sends nothing without a click
AI automationSales and recruiting outreach
2026 · in-house build

Manifest V3, and what it changes

Manifest V3 is the current platform for Chrome extensions, and Chrome has been phasing out the older Manifest V2. It changes how an extension is built, not only how it is packaged. These are the differences that decide an architecture, and how we handle each one.

AreaWhat Manifest V3 changedHow we build for it
Background workRuns in a service worker that Chrome stops when it is idleState lives in chrome.storage and timers in chrome.alarms, so nothing is lost when it sleeps
CodeEverything that executes ships in the package; remotely hosted code is not allowedBehaviour changes ship as reviewed updates; only data and settings load from a server
Network rulesdeclarativeNetRequest replaces blocking webRequestBlocking and redirect rules are declared up front, not run as code on every request
PermissionsHost access appears in the install prompt and in the store's reviewWe start from activeTab and optional host permissions, and widen only when a feature needs it
Page work in the backgroundA service worker has no DOMParsing and similar jobs move to an offscreen document
Where the interface livesPopup, side panel, options page, extension pages and content scriptsAI tools usually belong in the side panel, which stays open beside the page

Permissions deserve their own line. Every host permission appears in the install prompt and in the store’s review. We ask for the narrowest set that does the job: activeTab when the extension only acts after a click, optional host permissions requested at the moment a feature needs them, and access to every site only when the product genuinely works across every site. A short permission list is also easier to explain to the people being asked to install it.

If you already run an extension built on Manifest V2, it may have stopped working for your users. A migration is rarely a rename in the manifest. Long-lived background state has to move into storage, request blocking has to become declared rules, and any code loaded from a server has to come into the package. We read the old extension first and price the migration as a focused build.

AI Chrome extension development

Most AI extensions do one of three things. They summarise or answer questions about the page in front of the user, they rank or sort items the user has collected, or they draft text for the user to edit and send. The side panel is usually the right home for that work, because it stays open beside the page instead of vanishing the way a popup does.

Our own extension does the second and third. It ranks leads against a criterion the user types, sorts them into lists, and drafts the emails and the replies. A person reads every draft before anything leaves, which is the same rule our agentic AI development work follows: the AI does the reading and the drafting, and a person makes the call that cannot be undone.

Two decisions come before the first prompt. The first is where model calls go. API keys never ship inside an extension, because anyone can unpack an installed extension and read them, so the calls go through your backend or a small local helper. We use an OpenAI-compatible interface, which lets you change provider or run a local model without a rebuild.

The second is what leaves the browser. Sending a whole page to a model is easy and usually unnecessary. We send only the fields the task needs, write down what those are, and make the privacy policy say the same thing as the code. When the running cost matters, we estimate it per user before the build rather than after the first invoice.

Chrome Web Store review and privacy

The Chrome Web Store reviews every extension against its program policies before it goes live, and the same few things decide most outcomes. The extension needs a single purpose that is narrow and easy to understand. It may use the data it touches only for that purpose, which Google calls limited use. It needs a privacy policy that matches what the code actually does. And every permission needs a written justification in the developer dashboard, alongside a declaration of the data the extension handles.

We prepare the listing, the permission justifications and the privacy disclosures the review asks for, and submit under your own developer account so the listing belongs to you. Extensions that ask for broad host permissions get a closer review and can wait longer, which is one more reason to keep the list short.

Not every extension needs to be public. An internal tool can be unlisted, so only people with the link can install it. It can be published privately to your Google Workspace domain. Or it can be installed on managed Chrome browsers by policy, so it arrives on every work laptop without anyone searching for it.

Privacy work is not paperwork at the end. The single purpose and the data list are written in scoping, before the first line of code, and they are what keep the extension small. A feature that would need a new kind of data goes back through the same questions.

Edge, Firefox and Safari

Chrome extensions are built on the WebExtensions model that other browsers share, so one codebase can usually reach most of your users. Microsoft Edge runs on Chromium and takes the same package with little or no change, published through Edge Add-ons with its own listing and review.

Firefox supports Manifest V3 with differences that matter. Its background code runs as an event page rather than a service worker, and its APIs live under the browser namespace. A Firefox port is planned work, not a re-upload, and we keep the browser-specific parts in one place so the rest of the code stays shared. Safari needs the extension wrapped in a macOS or iOS app built in Xcode and released through the App Store, so we scope it as its own piece of work.

What an extension should never do

  • Run code fetched from a server. Everything that executes ships in the reviewed package.
  • Carry an API key or a secret. Anyone can unpack an installed extension and read it.
  • Ask for access to every site when one site will do.
  • Trust a message from a web page without checking where it came from.
  • Put sensitive data into the page, where the site's own scripts can read it.
  • Send page content anywhere the user did not agree to, or collect more than the task needs.
  • Leave a local helper open to the network. Ours listens only on 127.0.0.1 and checks the origin of every request.

An extension runs with more access than a web page, and once it is installed it stays quiet for years. That is why the rules on the left are not a checklist we run at the end. They decide the architecture: what lives in the content script, what lives in the background worker, and what never reaches the browser at all.

If you want an outside view of an extension, including one another team built, our cybersecurity service reviews the manifest, the permissions and the data flows, and tells you which findings matter for your users.

How an extension build runs

01 · day 0

Workflow call

Thirty minutes on the task, the sites it touches and who installs it. You leave with a fit answer and a rough estimate.

02 · scope

Purpose and permissions

The single purpose, the permission list, where AI calls go and what leaves the browser, written down with a fixed price.

03 · build

Builds you install

A new build on your team's own browsers every week, tested on the real sites it works on.

04 · release

Store or private release

The listing, justifications and privacy disclosures, or a private install by policy, then a documented handover.

What it costs

Focused extension
$3,500 to $7,500

One job, no backend of its own, about three weeks. The price includes the store listing or a private install.

Scoping
$2,500

Two weeks for a product-grade extension: the single purpose, the permission list, where AI calls go and a fixed estimate.

Prototype
$10,000 to $30,000

The extension working end to end with its backend, accounts and AI calls, in front of real users.

Upkeep
From $499 a month

Fixes when the sites it reads change, Chrome platform changes and store updates, after launch.

What moves a price within its range: how many sites the extension works on, whether it needs a backend and accounts, whether AI calls are involved, and whether it ships to one browser or several. When the extension is the front of a full SaaS product, the build follows our software ladder, where production runs $30,000 to $100,000. If your budget is nearer $10,000, affordable software development says what that buys. How we price.

Who this is for

A good fit

  • Sales and recruiting teams who spend the day in web apps they cannot change
  • Support teams copying between a helpdesk and a CRM
  • SaaS companies who want their product inside the tools their customers already use
  • Operations teams with one browser task that eats hours every week
  • Anyone whose Manifest V2 extension has stopped working

Probably not a fit

  • Auto-posting, auto-messaging or auto-applying bots
  • Collecting data at scale against a site's terms
  • Ad injection, link swapping or hidden tracking
  • Work that belongs in your own web app rather than someone else's page

If the task lives between systems rather than inside a tab, our AI automation service is usually the cheaper answer.

Common questions

A focused extension, one job with no backend of its own, costs $3,500 to $7,500 and is usually live in about three weeks. A product-grade extension with a backend, accounts and AI calls starts with two weeks of scoping for $2,500, then a prototype at $10,000 to $30,000. Upkeep after launch starts at $499 a month. When the extension is the front of a full SaaS product, the build follows our software ladder, where production runs $30,000 to $100,000.

Related

Have a browser task your team repeats every day?

Send the task and the sites it touches. Alex Novak, delivery director, replies within one business day with whether an extension is the right tool and a rough estimate.

Get an estimate