fix(mcp): reconnect to the browser after disconnect by yury-s · Pull Request #41966 · microsoft/playwright

Yury Semikhatsky (yury-s) · 2026-07-23T21:30:15Z

Summary

Recreate the MCP backend when its browser disconnected between tool calls, so the next call transparently reconnects: --extension reopens connect.html (silent with token bypass), cdp mode reconnects to the endpoint, persistent/isolated modes relaunch the browser.
Stop the single-shot extension relay and its http server once the browser connection drops.
Relaunch a crashed shared browser (--isolated/--shared-browser-context with multiple HTTP clients) instead of caching the dead one until all clients disconnect.

Dmitry Gozman (dgozman) · 2026-07-24T11:05:05Z

  });
 });

+test('transparently reconnects when the remote browser is restarted', async ({ cdpServer, startClient, server }) => {


Why is this a good thing? What if that's some completely unrelated browser? Do we have any evidence this is a useful scenario?

This is what has been asked for the extension. We discussed it offline ad decided to make it transpantly reconnect if the remote connection is lost for whatever reason. It especially makes sense when you connect from and mcp client over stdio which doesn't restart the server if it fails and would require manual intervention. When the server is reconnected the agent will see that the state has changed (due to a user action, crash, update or what not) and adapt.

Dmitry Gozman (dgozman) · 2026-07-24T11:06:46Z

+});
+
+test('http transport isolated closes the browser despite an earlier failed backend creation', async ({ serverEndpoint, server }, testInfo) => {
+  // A failed backend creation must not leak the client count, otherwise the


What does this comment mean?

There was a bug in program.ts, see new clientCount--; in the catch block. But I agree the comment does not make sense out of that context. Will remove.

Dmitry Gozman (dgozman) · 2026-07-24T11:08:25Z

  });
 });

+test('http transport isolated multiclient relaunches a crashed shared browser', async ({ serverEndpoint, server }, testInfo) => {


Is this a common scenario - with a crashed browser? I think this is more common when the user closes the browser window manually (for whatever reason) - do we want to relaunch it in this case?

Yes, I believe this makes sense as it is the mcp server that launched the browser. The same logic as above.

Dispose the backend when its browser disconnects between tool calls, so that the next call transparently creates a fresh one: --extension reopens connect.html, cdp mode reconnects to the endpoint, persistent/isolated modes relaunch the browser. Drop the cached shared browser when it disconnects and balance the client count when backend creation fails. Stop the single-shot extension relay and its http server once the browser connection drops. Fixes: microsoft#41874

…ulticlient modes

Drop the 'disposed' event and the disposeCallback parameter: the server resets its backend on 'disconnected', and program.ts tracks the context/browser teardown itself to tell a dead browser from a deliberate close.

github-actions · 2026-07-24T23:44:20Z

Test results for "MCP"

3 failed
❌ [firefox] › mcp/cli-core.spec.ts:179 › dialog-dismiss @mcp-windows-latest-firefox
❌ [firefox] › mcp/dashboard.spec.ts:108 › should activate session when show is called with -s @mcp-windows-latest-firefox
❌ [msedge] › mcp/cli-core.spec.ts:168 › dialog-accept @mcp-windows-latest-msedge

7845 passed, 1278 skipped

Merge workflow run.

github-actions · 2026-07-24T23:51:36Z

Hi, I'm the Playwright bot and I took a look at the CI failures on the latest run.

🟢 CI is clear — all three failures are pre-existing flakes

The three failures live in mcp/cli-core.spec.ts (dialog-dismiss, dialog-accept) and mcp/dashboard.spec.ts (session activation) — files this PR doesn't touch. Each one flips verdict on SHAs the PR can't be responsible for, so none is a regression.

Details

Overall: no failure is attributable to this PR. The diff is confined to the MCP backend reconnect path (browserBackend.ts, cdpRelay.ts, extensionContextFactory.ts, program.ts, server.ts) plus the launch/cdp/http/extension tests. The plain dialog-click and dashboard session-activation flows that failed never exercise a disconnect/reconnect.

Pre-existing flake / infra

[firefox] › mcp/cli-core.spec.ts:179 › dialog-dismiss (mcp-windows-latest-firefox) — pre-existing flake. Across the test-results DB firefox fails 5/1180 and passes the rest, including a failure on unrelated PR perf(test): avoid redundant stack capture #41961 (skn0tt-optimize-stack-capture, run 30017125515). This PR's reconnect change doesn't reach the dialog flow.
[msedge] › mcp/cli-core.spec.ts:168 › dialog-accept (mcp-windows-latest-msedge) — pre-existing flake. This exact msedge title also failed on unrelated PR fix(mcp): report malformed JSON config instead of falling back to INI #41909 (fix-41893, run 29863130236). The test is broadly flaky in this file: firefox 4/1180, msedge 2/279, chrome 1/1261 — failing across PRs fix(mcp): surface the missing executable path when a browser isn't installed #41941, feat(firefox): roll to r1539 #41970, chore(mcp): replace BackendManager with backend dispose callback and events #41982 too.
[firefox] › mcp/dashboard.spec.ts:108 › should activate session when show is called with -s (mcp-windows-latest-firefox) — pre-existing flake, the strongest signal of the three. Firefox fails 17/1180, and among those are 3 failures on main pushes and one on a release-1.62 push, plus a spread of unrelated PRs (feat(chromium): roll to r1235 #41974, fix: correct pending navigation log spacing #41949, feat(webkit): roll to r2339 #41975, test(chromium): fixme flaky "should ignore page.pause when headed" on macOS #41959, Revert "chore(deps-dev): bump fast-uri from 3.1.3 to 3.1.4" #41978, feat(test-runner): support named test locks #41862). Clearly independent of this PR.

_{Triaged by the Playwright bot - agent run}

Pavel Feldman (pavelfeldman)

Looks good to me, but Dima has questions in the tests.

## What's New ### Tool Improvements - **`browser_take_screenshot` — WebP format** — New `type` option to pick the image format (`png`, `jpeg` or `webp`); when unset, the format is inferred from the filename extension ([#41152](microsoft/playwright#41152)) - **`--codegen` — more languages** — Code generation now supports `python`, `java` and `csharp` in addition to `typescript` ([#42106](microsoft/playwright#42106)) - **Configurable settle delay** — New `--timeout-settle` flag (and `timeouts.settle` config option) to control how long to wait after each action for triggered work (navigations, requests) to settle, defaults to 500ms ([#41924](microsoft/playwright#41924)) ## Fixes - Aria snapshot capture is skipped when the response discards it ([#41923](microsoft/playwright#41923)) - New `--snapshot-boxes` flag (and `snapshot.boxes` config option) to enable bounding boxes in snapshots globally — previously this was only available per command ([#42102](microsoft/playwright#42102)) - Reconnect to the browser after a disconnect ([#41966](microsoft/playwright#41966)) - Report malformed JSON config instead of silently falling back to INI parsing ([#41909](microsoft/playwright#41909)) - Clear the dialog modal state when a dialog is closed out of band ([#42032](microsoft/playwright#42032)) - Detect downloads via the download event instead of guessing from `net::ERR_ABORTED` navigation errors ([#41933](microsoft/playwright#41933)) - Surface the missing executable path when a browser isn't installed ([#41941](microsoft/playwright#41941)) - Escape user input in codegen output ([#41962](microsoft/playwright#41962)) - Removed the stale `--output-mode` CLI option ([#41833](microsoft/playwright#41833)) ### Browser Extension - Validate `Host` and `Origin` headers on CDP relay WebSocket upgrades ([#42103](microsoft/playwright#42103)) - Launch the Chrome profile that has the extension installed ([#41939](microsoft/playwright#41939)) - Pass `noDefaults` for extension CDP connections ([#42119](microsoft/playwright#42119)) - Mention `PLAYWRIGHT_MCP_EXECUTABLE_PATH` in the error shown when the browser executable is not found ([#42122](microsoft/playwright#42122)) - Removed support for extension protocol v1 ([#41857](microsoft/playwright#41857))

Yury Semikhatsky (yury-s) requested a review from Pavel Feldman (pavelfeldman) July 23, 2026 21:30

This comment has been minimized.

Sign in to view

Yury Semikhatsky (yury-s) mentioned this pull request Jul 23, 2026

fix(mcp): auto-reconnect extension relay after disconnect #41875

Closed

This comment has been minimized.

Sign in to view

Dmitry Gozman (dgozman) reviewed Jul 24, 2026

View reviewed changes

Yury Semikhatsky (yury-s) requested a review from Dmitry Gozman (dgozman) July 24, 2026 17:11

This comment has been minimized.

Sign in to view

Yury Semikhatsky (yury-s) added 3 commits July 24, 2026 15:09

test(mcp): cover transparent browser relaunch for cdp, isolated and m…

b9b769f

…ulticlient modes

chore(mcp): signal backend teardown via 'disconnected' only

3761eda

Drop the 'disposed' event and the disposeCallback parameter: the server resets its backend on 'disconnected', and program.ts tracks the context/browser teardown itself to tell a dead browser from a deliberate close.

Yury Semikhatsky (yury-s) force-pushed the fix-41874 branch from 3c5757f to 3761eda Compare July 24, 2026 22:55

remove comments

b0abcfb

Harsh Mathur (harshmathurx) mentioned this pull request Jul 25, 2026

Chromium processes leak on MCP client disconnect (macOS, VSCode) #41918

Closed

Dmitry Gozman (dgozman) approved these changes Jul 27, 2026

View reviewed changes

This was referenced Jul 27, 2026

Revert "chore: roll stable test runner to 1.62.0-beta-1784842988000" #42003

Merged

fix(tsconfig): do not throw when "extends"/"references" cannot be resolved #42005

Merged

fix(types): support branded primitives in evaluate arguments #42009

Merged

Pavel Feldman (pavelfeldman) approved these changes Jul 27, 2026

View reviewed changes

Yury Semikhatsky (yury-s) merged commit 2cc9f3e into microsoft:main Aug 3, 2026
16 of 18 checks passed

Yury Semikhatsky (yury-s) deleted the fix-41874 branch August 3, 2026 16:40

Yury Semikhatsky (yury-s) mentioned this pull request Aug 5, 2026

chore: mark v0.1.18 microsoft/playwright-cli#443

Merged

This was referenced Aug 5, 2026

chore: mark v0.0.79 microsoft/playwright-mcp#1707

Merged

chore: mark v0.1.18 microsoft/playwright-cli#444

Merged

This was referenced Aug 6, 2026

Update Playwright MCP upstream to v0.0.79 swimmwatch/cloakbrowser-mcp#99

Open

Upstream updates detected — 2026-08-03 impravin22/my-claude-maestro#48

Open

Navigation Menu

Uh oh!

fix(mcp): reconnect to the browser after disconnect - #41966

fix(mcp): reconnect to the browser after disconnect#41966
Yury Semikhatsky (yury-s) merged 4 commits into
microsoft:mainfrom
yury-s:fix-41874

Yury Semikhatsky (yury-s) commented Jul 23, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

Dmitry Gozman (dgozman) Jul 24, 2026

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026

Uh oh!

Dmitry Gozman (dgozman) Jul 24, 2026

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026 •

edited

Loading

Uh oh!

Dmitry Gozman (dgozman) Jul 24, 2026

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

github-actions Bot commented Jul 24, 2026

Uh oh!

github-actions Bot commented Jul 24, 2026

Uh oh!

Pavel Feldman (pavelfeldman) left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

Yury Semikhatsky (yury-s) commented Jul 23, 2026

Summary

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

Dmitry Gozman (dgozman) Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

Dmitry Gozman (dgozman) Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Dmitry Gozman (dgozman) Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

Yury Semikhatsky (yury-s) Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

This comment has been minimized.

github-actions Bot commented Jul 24, 2026

Test results for "MCP"

Uh oh!

github-actions Bot commented Jul 24, 2026

🟢 CI is clear — all three failures are pre-existing flakes

Uh oh!

Pavel Feldman (pavelfeldman) left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Yury Semikhatsky (yury-s) Jul 24, 2026 •

edited

Loading