fix(mcp): do not close the browser twice when the client disconnects

[Pavel Feldman (pavelfeldman)]

Summary

• On stdin end, the MCP server closed its transport, which closed the browser, and the exit watchdog closed the browser at the same time. The second close force-killed the browser. Remove the 'end' handler, so the watchdog is the only shutdown path.
• The 'end' handler also killed a pending browser_annotate client. That client now notices the exit of its parent by itself: it reads stdin, so the 'close' event fires.

Fixes #43098

[github-actions]

Test results for "MCP"

1 failed
❌ [firefox] › mcp/idle-timeout.spec.ts:19 › closes the browser after the idle timeout and relaunches it on the next call @mcp-windows-latest-firefox

9047 passed, 1495 skipped

---

Merge workflow run.

[github-actions]

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

🟢 The remaining failure is a confirmed pre-existing flake

The PR is clear. The exact Windows Firefox test has failed on several unrelated PRs, before this change.

Details

Overall assessment

This PR changes shutdown after stdin disconnect and parent-process exit. The failing test keeps the MCP client connected and instead waits for the existing 500 ms browser idle timer, so those changed shutdown paths are not active. Its relaunch raced a browser close and returned Target page, context or browser has been closed.

Pre-existing flake / infra

• [firefox] › mcp/idle-timeout.spec.ts:19 › closes the browser after the idle timeout and relaunches it on the next call — confirmed pre-existing flake on mcp-windows-latest-firefox. The same test failed on unrelated test(mcp): work around Chromium losing WebMCP tools in same-origin iframes #42888, chore(coverage): hide public coverage tracing options before release #43030, chore: make check-deps work on Windows #43037, fix(cookies): clearCookies({ name: '' }) should only remove nameless cookies #43129, and fix(routeWebSocket): route WebSockets in already loaded documents #43130. The test(mcp): work around Chromium losing WebMCP tools in same-origin iframes #42888 triage found the same error on five other PRs and 6 failures in 229 aggregated runs (2.6%).

_{Triaged by the Playwright bot - agent run}