#Claude MCP Down Again?

1 messages · Page 1 of 1 (latest)

hardy dome
#

Hey @long wyvern , hitting a persistent 502 origin_bad_gateway on the Composio MCP endpoint (connect.composio.dev/mcp) for the last ~45 min. Every authenticated tool execution fails; it worked fine earlier in the same session.

What I've ruled out:

The endpoint is alive: a direct unauthenticated request to connect.composio.dev/mcp returns 401 (not 5xx).

Your status page says "all systems operational."
A different MCP server on the same client + proxy works fine right now, so the upstream proxy isn't broadly down.
Reconnecting / re-authing the Composio server did not fix it (fresh session, same 502).
So it looks isolated to the authenticated MCP gateway path for my connection.

Timeline: worked until ~08:55 UTC 2026-05-27, failing continuously since ~09:20 UTC. Happens even on a single-tool call (not batch size).

Error:
HTTP 502 Bad gateway
error_name: origin_bad_gateway
error_category: origin
detail: "The origin web server returned an invalid or incomplete response..."
retryable: true, retry_after: 60
Sample Cloudflare ray_ids (2026-05-27 UTC):
a02428fdfeb770af (09:58), a0242475bd9370af (09:55), a02414d7982470af (09:44), a0240ba01bfa5002 (09:38)

Anyone else seeing this?

scarlet dock
#

Same on my end, it is so annoying.

hardy dome
long wyvern
#

Hey folks, sorry for the disruption here. What you're hitting on connect.composio.dev/mcp ties back to Composio For You being temporarily down right now following our security incident. The 502s on the authenticated MCP gateway path are part of that.
We're working on getting things back up shortly. Subscribe to the status page for any updates. Appreciate the patience🙏🏼.

Composio
Composio - Status

Composio Status

hardy dome
royal plank
#

+1 on the 502s, identical signature here @hardy dome . Just completed fresh OAuth setup on connect.composio.dev/mcp with a valid JWT issued by login.composio.dev/oauth2/authorize. Every request returns consistent HTTP 502 origin_bad_gateway from Cloudflare (5/5 attempts).

@long wyvern thanks for the acknowledgment here. Separate to the 502, I have a much larger compounding issue from the same security incident that I have emailed support about 3 times since 25 May with no reply: across 3 of my Composio accounts, all of my original projects, project API keys, and ~11 MCP servers are no longer visible. Each account now shows a single auto-generated "_first_project" workspace dated 24 May with zero API keys and zero MCP servers. The old API keys returned "API key is not allowed to access this MCP server" (not "invalid") before I rotated them, which suggests the servers still exist server-side but in projects I can no longer access.

Once Composio For You comes back up: will my original projects + servers be there, or were they destroyed in the breach response? This determines whether I wait or sink days into an OAuth rebuild across 4 accounts. Would really appreciate a clear yes/no rather than another email going unanswered.

halcyon lake
#

Same issue here... Annoying that status page does not reflect it either. I trust you guys are working hard to fix, so thanks! Please keep us updated

long wyvern
#

Sorry for the status page and quick update — the Connect MCP (connect.composio.dev/mcp) is back up and fully operational. You can wire it up in your client (Claude, Cursor, n8n, etc.) and connect any apps you need from there; on Claude, disconnecting and reconnecting the connector usually does the trick. Certain write operations on the For You UI are still temporarily limited as a precautionary measure following our security incident — going through Connect MCP is the cleanest path while that's the case. Lmk how it goes.

halcyon lake
halcyon lake
long wyvern
#

@halcyon lake, could you try add this server as a connector directly and lmk if it works for you?