Summary
In MCPClientManager, the connect/restore path that calls discoverIfConnected(serverId) reports a failed discovery with console.error, including the recoverable case where the result is Discovery timed out after 15000ms. Fleet alerting keyed on error-level logs pages on that line even though the caller can and does recover by explicitly rediscovering the server afterwards.
Seen on agents@0.23.0.
Suggestion
- Log the discovery timeout at
warn severity rather than error.
- Emit an observability event for it so callers can react without scraping console output, e.g.
_onObservabilityEvent.fire({ type: "mcp:client:discover", payload: { serverId, error }, timestamp }).
We carry both as a local patch (warn plus the event for the timeout case, console.error retained for other discovery failures). Happy to open a PR if that split is acceptable.
Summary
In
MCPClientManager, the connect/restore path that callsdiscoverIfConnected(serverId)reports a failed discovery withconsole.error, including the recoverable case where the result isDiscovery timed out after 15000ms. Fleet alerting keyed on error-level logs pages on that line even though the caller can and does recover by explicitly rediscovering the server afterwards.Seen on
agents@0.23.0.Suggestion
warnseverity rather thanerror._onObservabilityEvent.fire({ type: "mcp:client:discover", payload: { serverId, error }, timestamp }).We carry both as a local patch (warn plus the event for the timeout case,
console.errorretained for other discovery failures). Happy to open a PR if that split is acceptable.