Sustained 502 Bad Gateway from mcp-proxy.anthropic.com (Gmail MCP)

Confirming this from a second account/region — same failure mode, broader diagnostic context than originally captured.

Environment:

Claude Code CLI v2.1.109 (same as reporter)
Windows 11 Home 10.0.26200 (VS Code extension)
Region: Perth, Australia (APAC — same region family as the Melbourne edge cited)
Organization ID: 01fbe21b-c66d-49a1-9180-84fe7f3c0dd3
claude doctor clean, on latest, auto-updates enabled
Additional Cloudflare Ray IDs (all 2026-04-15 UTC):

9ec85754cf80120c — 04:52:34
9ec8578c583e120c — 04:52:43
9ec8632c9cea8ad0 — 05:00:40
9ec87683ffb9cffa — 05:13:52
9ec876c8ed4bcffa — 05:14:03
9ec8cc701badafce — 06:12:31
Pattern observed — payload/scope sensitivity:

search_threads with newer_than:1d, pageSize:5 → succeeds (intermittently)
search_threads with newer_than:7d, pageSize:50 → consistently 502s
search_threads with newer_than:7d is:unread, pageSize:10 → succeeds Suggests the proxy fails under larger result payloads / longer backend processing, not on scope alone.
Key isolating diagnostic: the same claude.ai-managed Gmail connector works correctly when invoked from claude.ai web chat against the same account. The failure is specifically in the Claude Code → mcp-proxy.anthropic.com route — not the connector, not Google OAuth, not Gmail API.

Duration: started ~13 April 2026 for my account; Gmail MCP via Claude Code worked successfully on 13 April earlier in the day, then began 502-ing shortly after. 2+ days of sustained failure.

Related but distinct from #42509 (that issue concerns custom FastMCP/Tailscale servers; this one is the first-party hosted Gmail MCP).

My parallel report at #48291 can close as a dup of this — linking here for engineering cross-reference.

I'm hitting the same issue. A few additional data points:

▎ - First Gmail MCP call per session succeeds. Every subsequent call 502s. Consistent
pattern across every session for 3 days (Apr 13–16).
▎ - Other MCP servers on the same mcp-proxy.anthropic.com host work fine — Fireflies
handles dozens of calls per session without issues.
▎ - Google Calendar MCP also works fine — so this isn't a general Google OAuth issue, it's
specific to the Gmail connector.
▎ - macOS, Claude Code CLI, Anthropic API.

Yep, same here for about 24 hours now. Gmail MCP via mcp-proxy.anthropic.com has been returning 502 consistently. list_labels worked once, search_threads 502'd every time. Different Cloudflare edge (Chicago vs. Melbourne), so probably global.

I'm using Opus 4.6 via OpenRouter in OpenCode - not sure who it's routing to but I've been getting 502s all day.

Re-added Gmail MCP up at .claude/settings.json and this resolved for me.

Thanks for the report. Claude Code was reaching the wrong upstream for the Gmail connector while claude.ai web reached the right one, which is why you saw 502s in one and not the other. That was corrected on April 18 and a commenter confirmed working on April 24. Closing this; please reopen if the 502s come back on a current build.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.