fix(mcp): report malformed JSON config instead of falling back to INI

[dcrousso]

loadConfig fell back to INI parsing on any JSON error, so a typo'd JSON config (e.g. a trailing comma) was silently accepted as INI: the intended settings were dropped for the defaults and the raw JSON kept as an unused top-level key

surface the original SyntaxError when the input is a JSON object ({) while keeping the INI fallback (including files whose first line is a [section] header) and BOM handling

fixes #41893

[yury-s]

Maybe it's time to make .ini support official or drop it? @pavelfeldman wdyt?

[github-actions]

Test results for "MCP"

2 failed
❌ [chrome] › mcp/cli-navigation.spec.ts:56 › goto chrome:// page that closes the tab does not crash the response @mcp-windows-latest-chrome
❌ [msedge] › mcp/cli-core.spec.ts:168 › dialog-accept @mcp-windows-latest-msedge

7790 passed, 1269 skipped

---

Merge workflow run.

[github-actions]

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

🟡 Uncertain, but both failures lean hard toward pre-existing flakes

Two Windows-only MCP failures, neither on the config-parsing path this PR touches. One is a known pre-existing flake; the other I can't stamp outright because its only failure in the results DB is this PR's own run.

Details

This PR changes only loadConfig's JSON/INI handling in packages/playwright-core/src/tools/mcp/config.ts (plus tests/mcp/config-resolve.spec.ts). Both failures come from the same test run 29863130236 on Windows, and neither test loads a config file — so the diff can't reach either one.

Pre-existing flake / infra

• [chrome] › mcp/cli-navigation.spec.ts:56 › goto chrome:// page that closes the tab does not crash the response (@mcp-windows-latest-chrome) — confirmed flake. Across the results DB this test fails 16 of 117 runs (14%) on mcp-windows-latest-chrome and is green on every other bot/OS. Crucially it fails on runs the PR can't be responsible for: main pushes (e.g. 29855614143, 29845175997) and unrelated PRs (move-skills, expect-fn, roll-into-pw-webkit/2336, fix-21484). Same toContain signature every time, only on Windows chrome — a timing flake, not this diff.

Uncertain (almost certainly a flake)

• [msedge] › mcp/cli-core.spec.ts:168 › dialog-accept (@mcp-windows-latest-msedge) — a toContain mismatch on the alert-dialog output. Passes 116 of 117 runs on this bot and 100% on every other bot/browser; the lone failure in the DB is this PR's own run (29863130236, sha 7faf1c08). Because there's no PR-independent failure to point to, I can't satisfy the "same test failing where the PR can't be responsible" rule and call it a flake outright — but the changed code (JSON/INI config parsing) has nothing to do with clicking a button and accepting a dialog, and this is Windows-msedge only.

Caused by this PR

• None. No failure lands on the config-loading path the PR changes.

To close out the uncertain one, a green re-run of the current head (08137e37) on mcp-windows-latest-msedge would do it.

_{Triaged by the Playwright bot - agent run}