MV2-to-MV3 Side-by-Side Replay
When migrating an extension to MV3, replay real browser flows side by side to pinpoint the exact step where the new version loses behavior the old version had.
After Chrome removed the installation channel for legacy MV2 extensions, maintainers are less worried about whether they can rewrite an API than whether an MV3 release quietly stops blocking a class of requests on a real page. MV2 and MV3 are two generations of Chrome extension rules. Teams upload the old package still in production and the new package awaiting release, then select the pages, login states, and click flows that best represent real user behavior.
The product launches two isolated browsers and has the old and new versions shadow-run the same sequence: opening tabs, switching accounts, visiting dynamic pages, triggering network requests, and changing permission states. It aligns page outputs, request logs, storage changes, and console output from both sides. When results diverge, the report stops at the first differing action and includes the trigger conditions, relevant APIs, and a runnable minimal reproduction script.
Maintainers can save a flow after a fix as a regression case and rerun it automatically on code submission. The first release focuses on network requests and background tasks common to ad blockers, script managers, and privacy extensions; it does not attempt to design a team’s entire migration plan. Its output is a list of behavioral differences that engineers can immediately reproduce and fix.
Why now
On August 31, 2026, the Chrome Web Store removed all remaining MV2 extensions. S1 Installed legacy versions can no longer be updated or reinstalled, making maintainers more urgent to catch silent MV3 behavior regressions before release. S1
Target user
Small teams still maintaining ad-blocking, script-management, or privacy extensions. It is most useful just before an MV3 candidate package is submitted to the Chrome Web Store, or immediately after a migration fix is merged. At that point, unit tests usually pass, but teams worry about behavioral drift after login-state changes, on dynamic pages, or during permission changes. The on-call maintainer needs differences they can reproduce directly, not another migration checklist.
Minimal entry point
Use Playwright to control two isolated persistent browser profiles, loading one unpacked package into each. S2 The initial dual-run setup pins both to the same Chrome 138 baseline and enables the official retention policy. S1 Recorded actions cover only navigation, clicks, account switching, and permission changes. A collection layer uses Playwright events and CDP to capture DOM summaries, requests, console output, and storage snapshots. The aligner compares at action boundaries rather than seeking pixel-perfect page matches. Reports lead with the first divergence, then generate a rerunnable Playwright case. Once the versions match, the MV3 package gets a standalone smoke test in current Chromium.
Punching above its weight
The first users are extension repositories on GitHub that still maintain both MV2 and MV3 builds. Package the runner as a reusable GitHub Action that attaches the first behavioral divergence and a reproduction script to migration PRs. Select public ad-blocking or script-management projects and submit verifiable failing cases. Then publish a set of redacted difference reports so maintainers can judge whether it saves more investigation time than their existing E2E tests.
Competitors & gaps
- Custom Playwright extension E2EGoogle
- Playwright can already load extensions in persistent Chromium contexts and access MV3 service workers. S2 Teams can script logins, clicks, pop-ups, and page assertions, and plug them into existing CI. It suits engineering teams that already know what they need to verify. But test authors must still write expected outcomes and maintain environments for both the old and new packages. It does not automatically align requests, storage changes, and console logs across the two versions by action. When dynamic pages diverge, teams must still determine where the first causal divergence occurred. Scenarios such as service-worker suspension and permission changes also require fixtures tailored to the extension’s architecture. The opportunity is to make dual-version orchestration, noise reduction, and minimal-reproduction generation the default workflow.
- BrowserStack AutomateGoogle
- BrowserStack Automate supports uploading Chrome extension packages and running them in remote Playwright sessions. S3 It can also retain network logs and session results. S3 Its browser infrastructure, platform matrix, and team collaboration are mature. For cross-environment compatibility testing, it is more convenient than operating an in-house browser pool. Its existing workflow still centers on test scripts driving one extension package, with assertions written in the script. Its documentation does not provide synchronized MV2-versus-MV3 replay, state alignment, or first-divergence detection. S3 Teams must still handle matching login states, page data, and noise reduction themselves. This product could serve as a migration-diagnostics layer on top, or use local runners for sensitive extensions and accounts.
How it makes money
Charge per team on a subscription that includes a fixed allowance of parallel replays and CI runs. Bill overages by browser runtime. Offer local deployment and private runners at higher tiers.
The case against
Dynamic sites can create a high volume of false differences. Ad rotation, A/B tests, timestamps, and API responses may diverge before the extension does. Normalize too aggressively and genuine missed blocking may disappear. Login flows also involve CAPTCHAs, two-factor authentication, and expired sessions, so replay stability directly affects report credibility. The legacy runner must remain pinned to Chrome 138; maintaining an outdated browser and a high-risk sandbox is not easy over time. S1 Extension packages and login states are sensitive as well, so isolation, key destruction, and local deployment raise delivery costs. If the first-divergence report produces frequent false positives, engineers will quickly return to manual retesting.