fix(mcp): do not ask for roots without the event stream

[Yury Semikhatsky (yury-s)]

Summary

• Server-to-client requests go over the standalone GET event stream. A streamable HTTP client that advertises roots but never opens that stream can't be delivered roots/list, so the request was dropped by the transport and the first tools/call stalled for the SDK's 60s default timeout.
• Skip listRoots() when the transport never became bidirectionally ready, instead of sending a request that could only time out.

Fixes #42256

[Yury Semikhatsky (yury-s)]

If the client advertises capabilities?.roots => both server->client transport and list/roots requests should work.

[github-actions]

Test results for "tests 1"

19 flaky

⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@frozen-time-library-chromium-linux`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@frozen-time-library-chromium-linux`
⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/popup.spec.ts:260 › should not throw when click closes popup `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@realtime-time-library-chromium-linux`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@realtime-time-library-chromium-linux`
⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@chromium-ubuntu-22.04-node20`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@chromium-ubuntu-22.04-node20`
⚠️ [chromium-library] › library/beforeunload.spec.ts:130 › should support dismissing the dialog multiple times `@chromium-ubuntu-22.04-node22`
⚠️ [chromium-library] › library/global-fetch.spec.ts:293 › should return security details from response `@chromium-ubuntu-22.04-node22`
⚠️ [chromium-library] › library/har.spec.ts:639 › should have security details `@chromium-ubuntu-22.04-node22`
⚠️ [firefox-library] › library/global-fetch.spec.ts:293 › should return security details from response `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/har.spec.ts:639 › should have security details `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-page] › page/page-emulate-media.spec.ts:144 › should keep reduced motion and color emulation after reload `@firefox-ubuntu-22.04-node20`
⚠️ [webkit-library] › library/global-fetch.spec.ts:293 › should return security details from response `@webkit-ubuntu-22.04-node20`
⚠️ [webkit-library] › library/har.spec.ts:639 › should have security details `@webkit-ubuntu-22.04-node20`

51183 passed, 1233 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

1 failed
❌ [chrome] › mcp/config.spec.ts:203 › chromiumSandbox › --sandbox enables the sandbox @mcp-ubuntu-latest-chrome

8122 passed, 1341 skipped

---

Merge workflow run.

[github-actions]

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

🟢 CI is clear — the one failure is a pre-existing flake

[chrome] › mcp/config.spec.ts:203 › --sandbox enables the sandbox fails on pushes to main and on unrelated PRs with the same signature, so it isn't caused by this PR. The "tests 1" report has no real failures (19 flaky, 0 failed).

Details

Overall: This PR changes MCP roots/list handling (server.ts) and trace-viewer JSON pretty-printing (networkResourceDetails.tsx). Neither touches browser channel selection or sandbox launch, which is what the failing test asserts on. No group is attributable to this PR.

Pre-existing flake / infra

• [chrome] › mcp/config.spec.ts:203 › chromiumSandbox › --sandbox enables the sandbox — pre-existing flake, error SyntaxError: "undefined" is not valid JSON (i.e. browser_get_config returned no result). Across the test-results DB this chrome-project test produces a failed result in essentially every recent run, including pushes to main (SHAs be6e7cc9, 51023b42, b856094e, 5f8e7eac) and unrelated PRs feat(browserContext): add windowState option #42286, fix(mcp): do not clobber chromiumSandbox from the config file #42288, chore(html-reporter): render a thumbnail for every trace attachment #42289, fix(dashboard): don't strand a stale tab selection on session switch #42290, fix(chromium): stop disabling BoundaryEventDispatchTracksNodeRemoval #42294, fix(trace-viewer): pretty-print JSON without losing precision #42297 — all with the identical signature. It's usually rescued on retry; on this run it wasn't. Since it fails where this PR can't be responsible, it's not caused by the change.

Not failures

• The 19 entries in the "tests 1" report (global-fetch.spec.ts:293, har.spec.ts:639 security details, popup.spec.ts:260, beforeunload.spec.ts:130, page-emulate-media.spec.ts:144) are all flaky (passed on retry), across chromium/firefox/webkit — nothing to triage, and none overlap the MCP/trace-viewer code this PR touches.

_{Triaged by the Playwright bot - agent run}