Bug fix: recover browser MCP calls after transport failures (macOS)

[swordfish444]

Bug Fix Summary

This PR fixes a reliability bug in Codex browser MCP handling on macOS.

Linked bug: #13138

Bug Description

When browser MCP transports die mid-session, follow-up tool calls can fail with transport-level errors (Transport send error, Broken pipe, Transport closed), and browser MCP workflows become unreliable until restart.

Affected MCP servers:

• chrome-devtools
• playwright

Fix

• Keep one-shot restart + retry behavior in McpConnectionManager
• Expand transport-failure matching to include:
  • transport closed
  • transport send error
  • transport receive error
  • transport errors containing broken pipe / connection reset
• Add targeted unit coverage for matcher behavior

Validation

Automated

• just fmt
• just fix -p codex-core
• cargo test -p codex-core is_transport_closed_error
• cargo test -p codex-core mcp_tool_call_recovers_from_transport_closed

End-to-end (macOS, patched app-server)

Server:

• ./target/debug/codex app-server --listen ws://127.0.0.1:4322

Chrome DevTools MCP:

1. first turn tool call succeeds (mcp__chrome-devtools__new_page)
2. force-close child MCP process (kill -9 10162)
3. second turn in same thread succeeds without app restart
4. server log confirms recovery path:
   • MCP transport closed for server chrome-devtools; restarting

Playwright MCP:

1. first turn tool call succeeds (mcp__playwright__browser_tabs)
2. force-close child MCP process (kill -9 10154 11782)
3. second turn in same thread succeeds without app restart
4. server log confirms recovery path:
   • MCP transport closed for server playwright; restarting

Related prior report: #6649

[swordfish444]

Tagging additional maintainers for this bug fix: @etraut-openai @mzeng-openai. This PR is a reliability bug fix (linked to #13138) with macOS E2E validation for browser MCP transport drops (Chrome DevTools + Playwright), including forced process-kill recovery in-session without restarting Codex.

[etraut-openai]

@swordfish444 Please do not submit PRs. If you want to suggest a new feature, file a feature request.

[swordfish444]

Tracking bug report (per maintainer guidance): #13138

[swordfish444]

@etraut-openai Thanks for the guidance. I opened #13138 as requested. To clarify, this is not a feature request; it is a reliability bug affecting existing behavior in Codex on macOS where browser MCPs (chrome-devtools and playwright) can fail with transport errors (including broken pipe / transport closed) and become unusable without restart.

Given the user impact and the fix + E2E validation already captured in this PR, I’d appreciate it if you could reopen #13100 for review.

[swordfish444]

Issue mapping (to avoid duplicate confusion):

• Fix target here: app/browser MCP transport-drop recovery for Chrome DevTools + Playwright (tracked in Bug: browser MCP transports drop on macOS and require app restart (Chrome DevTools + Playwright) #13138).
• Related umbrella: MCP client does not auto-reconnect after disconnect #11489 (generic MCP reconnect gap).
• Distinct case: CLI degrades after stale streamable-http MCP session; subsequent tool calls fail #12869 (CLI stale streamable-http session / 401), different repro path.

This PR should be evaluated as a reliability bug fix for existing behavior, not a feature request.