feat(mcp): allow saving browser_find results to a file

[Yury Semikhatsky (yury-s)]

Summary

• Add filename option to browser_find and --filename to the CLI find command, so large result sets can be written to a file instead of the response.

References #42077

[Pavel Feldman (pavelfeldman)]

suggestedFilename: params.filename - wrong?

[github-actions]

Test results for "MCP"

1 failed
❌ [webkit] › mcp/annotate.spec.ts:417 › should cancel browser_annotate when the MCP client disconnects @mcp-windows-latest-webkit

8789 passed, 1480 skipped

---

Merge workflow run.

[github-actions]

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

🟡 One failure, very likely unrelated, but not proven a flake on WebKit

The only failure is mcp/annotate.spec.ts:417 on Windows WebKit. This PR only changes browser_find, so it doesn't touch annotate. The same test flakes on other Windows browsers, including on main, but I found no earlier WebKit failure.

Details

Overall: this PR adds an option to save browser_find results to a file. It touches tools/backend/find.ts, the CLI daemon commands.ts, the skill doc and the find specs. None of that runs in the annotate/dashboard flow. The MCP run (36454676012) had 8789 tests pass and 1 fail.

Uncertain

• [webkit] › mcp/annotate.spec.ts:417 › should cancel browser_annotate when the MCP client disconnects (mcp-windows-latest-webkit). The diff doesn't reach this test. The test-results DB (latest data from 2026-09-25) shows the same test flaking only on the Windows MCP bots, on runs this PR can't be responsible for:

  • firefox: 7 of 705 runs failed
  • chrome: 3 of 718
  • msedge: 3 of 716
  • chromium: 1 of 720

  Some of those failures are on main itself (e.g. runs 34246196589, 33888071298, 33661279401, 32276523116). The others are on unrelated PRs such as chore(proxy): drop legacy 'per-context' launch proxy handling #42892, feat(video): allow recording video at a custom fps #42752 and fix(webkit): do not report blob: URLs to route handlers #42738. Every failure has the same signature: getByRole('main', { name: 'Dashboard: annotate' }) never becomes visible.

  The webkit project on this bot has 0 failures in 727 recorded runs, so there's no history of this exact (project, test) pair failing. That's why this is yellow and not green. I couldn't get this run's artifacts, so I haven't confirmed that the error here matches the Dashboard-not-visible signature. If it does, this is almost certainly the same Windows flake. Re-running the job should clear it.

_{Triaged by the Playwright bot - agent run}