Claude Code Retries Dropped Remote MCP Servers in the Background for About a Day

Claude CodeView original changelog

Claude Code 2.1.327 fixes remote MCP servers staying disconnected after a longer outage such as a VPN drop. Most servers on other machines reached by hostname over https are now retried in the background for about a day, so tools come back without a manual reconnect.

Key Takeaways

  • Remote MCP servers now reconnect on their own after a long outage such as a VPN drop.
  • Retries run in the background for about a day.
  • The behaviour applies to most https servers reached by hostname on other machines.
  • Users no longer need to reconnect manually after a network blip.
  • /mcp now explains next steps when it shows no servers.
  • Long-lived sessions become more dependable on flaky networks.

Reconnecting after a long outage

Before 2.1.327, a remote MCP server that dropped during a longer outage, such as a VPN disconnect, could stay disconnected for the rest of the session. The user had to notice that its tools were missing and reconnect by hand.

Claude Code now retries most servers on other machines that are reached by hostname over https in the background for about a day. When the network returns, the connection and its tools come back on their own. Servers that are local or reached another way are not covered by this behaviour.

Related MCP fixes

The /mcp screen also now says what to do when it reports no servers after one was added during a session or declined, instead of only saying none were configured.