Gmail / Calendar hosted MCP endpoints (*.mcp.claude.com) return 404 — deprovisioned or outage?
Summary
The hosted MCP endpoints https://gmail.mcp.claude.com/mcp and https://gcal.mcp.claude.com/mcp return HTTP 404 from Google's frontend for every path, including a properly-formed MCP initialize request. Trying to adopt them via claude mcp add produces ✗ Failed to connect, and /mcp → authenticate fails (previously with a TLS error during the OAuth init, now 404).
Seeking confirmation: were these retired, or is this an outage? If retired, is there a replacement path for Gmail/Calendar in Claude Code?
Reproduction
$ claude mcp add --transport http --scope user gmail https://gmail.mcp.claude.com/mcp
$ claude mcp list
gmail: https://gmail.mcp.claude.com/mcp (HTTP) - ✗ Failed to connect
gcal: https://gcal.mcp.claude.com/mcp (HTTP) - ✗ Failed to connect
Protocol-level probe:
$ curl -s -X POST https://gmail.mcp.claude.com/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "MCP-Protocol-Version: 2025-06-18" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize", ...}'
HTTP/2 404
server: Google Frontend
<html>...Error: Not Found...The requested URL /mcp was not found on this server...</html>
Compare against a known-working HTTP MCP:
$ curl -s -X POST https://mcp.slack.com/mcp (same headers)
HTTP/2 401
server: Apache
www-authenticate: Bearer resource_metadata=\"https://mcp.slack.com/.well-known/oauth-protected-resource\"
Slack's MCP returns a proper Bearer challenge; *.mcp.claude.com returns a Google generic 404, suggesting no backend is routed behind the hostname.
Evidence the endpoints used to work
Internet Archive Wayback Machine captured https://gmail.mcp.claude.com/ and https://gcal.mcp.claude.com/ returning 200 OK HTML (titled "Claude for Gmail", with claude_logo.png) as recently as April 13, 2026 for both hosts. Dozens of snapshots from Feb–Mar 2026 show consistent serving.
Other users reported successful installs in late March 2026 in community channels. Multiple community users confirm they worked "previously" but are broken now.
Environment
- Claude Code 2.1.126 (via Bedrock)
- macOS 15.3
- User-scope MCP registration
What I've already checked
- Not a TLS/cert issue:
curlhandshake validates cleanly with and without a custom CA bundle - Not a probe-format issue: proper
initializePOST with protocol headers also returns 404 - Both sibling hosts (
gmail,gcal) broke simultaneously — coordinated, not isolated failure - No retirement notice in the Claude Code CHANGELOG, docs.claude.com, or anthropic.com/news
- Only reference in official docs treats Gmail/Calendar as claude.ai connectors (web UI), not as Claude Code hosted MCPs — so these may have always been an undocumented preview
Questions
- Were these endpoints intentionally retired? If so, could the retirement be noted in the CHANGELOG so other users stop chasing TLS/config red herrings?
- Is there a replacement path for Gmail + Calendar in Claude Code that doesn't require users to provision their own GCP OAuth client? (Many users lack GCP project-creation rights in their corporate Workspace orgs and relied on these hosted connectors specifically for that reason.)
- If the connection exists within claude.ai, is exposing it to Claude Code on the roadmap?
6 Comments
Thank you for the report, @tomra-hub - you can use the built-in connectors in Claude to connect to Gmail and Google Calendar.
This is still an issue. We need it fixed before we can move forward with the managed agents.
I spend probably 8 hours in the last 24 fighthing this issue before I found this thread.
I'm also getting the same issue with managed agents. The addition of the credential within the vault reflects the correct URL but when the credential is stored the incorrect/dead URL is stored alongside it so Managed Agents are not able to access the MCP since the credential auth is locked to the dead URL.
@localden this issue is not solved at all, google calendar mcp connection in Claude Console for Managed Agent does not work. Please re-open and fix it.
I'm also experiencing this in Managed Agents so if someone has a fix that would be very helpful. I see the last comment was 3 weeks ago. Thanks!
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.