Repository navigation
MCP: connections remain failed after laptop sleep/wake #57015
Description
Description
I closed my laptop while an MCP request was in progress. After waking it and restoring network access, several MCP servers showed Error and did not automatically reconnect.
For cloudflare-builds, the details showed:
Status: failed
Configuration: mcp.servers.cloudflare-builds
Error: Request timed out
I expect enabled MCP servers to recover from temporary connection loss without reconnecting each one manually.
Could OpenCode retry failed connections with backoff after sleep or network interruption? Disabled or manually disconnected servers should stay disconnected. Recovery should not automatically replay interrupted tool calls, since they might already have completed.
Steps to reproduce
- Connect to a remote MCP server.
- Start an MCP request.
- Close the laptop while the request is active.
- Wake the laptop and restore network access.
- Observe the failed MCP connection.
This describes the observed sequence; repeatability has not yet been checked.
OpenCode version
v2.0.32
Operating System
macOS
Terminal
iTerm2
Reacted by Yuji Hatakeyama
Activity
opencode-agent commentedon Oct 19, 2026
Thanks for the clear report! I think this is a duplicate of #52237: remote MCP servers on macOS end up
failed: Request timed outafter sleep/wake and are never retried until the service restarts.Possible duplicates and closely related items, ordered from most to least likely:
Request timed out, no retry); asks for reconnect with backoffErroruntil the server process is killedUntil that's resolved, you can reconnect a server manually from
/mcps, or runopencode service restartto bring them all back.The bot will close this issue as a duplicate in 1 day unless you reply explaining how your problem is different.
opencode-agent commentedon Oct 20, 2026
Closing this as a duplicate of #52237. That issue covers the same problem: remote MCP servers stay
failed: Request timed outafter sleep/wake and are never retried. We waited a day in case your case was different, and there's been no reply, so please follow #52237 for updates. Open PR #54378 adds reconnect with backoff for failed servers. Until then, you can reconnect a server from/mcpsor runopencode service restart. If your problem turns out to be different, reply here and we can reopen.