Resolved -
Between approximately 01:30 and 04:45 UTC on July 8, one of our backend application servers lost its network connection to a supporting datastore.
Requests that happened to route to that one server and triggered background work either failed or responded slowly during that window; our other servers were unaffected, which is why the impact was intermittent rather than widespread. We deployed connection hardening that makes this path more resilient, and the errors stopped immediately and have not recurred.
The MCP gateway and agent-to-tool traffic are fully operational, and we found no data loss. We're sorry for the disruption.
Jul 8, 14:08 EDT
Monitoring -
We've deployed the fix for the intermittent write failures and slow responses seen during overnight windows on July 7–8. The affected connection now enforces a strict timeout and reconnects automatically, so a brief network interruption fails in seconds and background work continues instead of stalling.
Write operations and response times should now behave normally, with no action needed on your end. We're monitoring to confirm the fix holds before we mark this resolved.
Jul 8, 00:49 EDT
Identified -
We've identified the cause of the intermittent errors some customers saw on save and update operations today (July 7). During short windows, certain write actions failed on the first attempt and succeeded on retry; the service otherwise stayed available throughout, and no data was lost.
The errors trace to brief, momentary interruptions in the network connection between our application and its backend datastore.
We are hardening our connections now, so a brief interruption self-heals rather than failing an operation. We're also confirming with our datastore provider whether any network events occurred on our side.
We'll post another update as our mitigation lands.
Jul 7, 23:10 EDT
Investigating -
We are investigating a disruption affecting the MCP Manager web application. Availability is currently intermittent — the app is loading for some requests and failing to load for others. Engineers are actively engaged and this is our top priority. We are working to determine the cause and restore consistent access, and we'll post another update shortly.
Jul 7, 22:25 EDT