Make the MCP Stack Self-Healing

QUESTNET // PROJECT CHRONICLEFILE 2999
In progressDifficulty: HardEst. cost: Free CADOwner: Brent & AriUpdated 2026-11-02Time: One focused infrastructure session
Brent and Ari in their neon-lit lab designing a self-healing MCP stack with automatic startup, watchdog monitoring, restart recovery, and service health checks, while Tucker and Shyshy sleep nearby.

Harden our MCP infrastructure so every MCP component starts automatically when its host starts and is supervised after launch.

The quest covers identifying each MCP service and its host, giving every component a reliable auto-start mechanism, adding watchdog or restart-on-failure behavior, preventing duplicate instances, capturing useful logs, and verifying recovery after both deliberate process termination and a full host reboot.

Completion criterion: all MCP components recover without manual babysitting after normal restarts and unexpected process failures, and we can tell from logs/status when a component has failed or restarted.

Verified milestones from the former Quest Log

October 5, 2026: Ari-Core MCP self-healing proven

The Ari-Core/Linux filesystem MCP tunnel was configured as an enabled Brent-user systemd service with lingering, Restart=always, and a 10-second restart delay. Forced tunnel termination resulted in a new tunnel process and MCP child, with health and readiness checks passing. After a complete Debian reboot, with the machine left at the login screen, the lingering user manager restored the tunnel and filesystem MCP; remote access succeeded without manual login or SSH.

The power-outage incident revealed an outstanding hardware prerequisite: Ari-Core’s RTC/CMOS settings were not being retained, requiring firmware intervention to restore the Other OS setting before Debian could boot. Unattended cold-power-loss recovery therefore remained unresolved.

October 5, 2026: AriVision Windows recovery proven

The Windows-side AriVision MCP stack was configured with a user-encrypted control-plane credential, a supervisor for its local MCP server and tunnel client, and a scheduled one-minute heartbeat with MultipleInstances IgnoreNew. Testing exposed a weakness in Windows Task Scheduler restart behavior: after supervisor termination, orphaned children remained and the task showed Ready. The startup sequence was hardened to clean up orphaned AriVision processes. A second supervisor-death test successfully rebuilt the server and tunnel and restored remote access.

BrentPuter was subsequently rebooted. After Brent logged into Windows, the supervisor and AriVision MCP stack recovered without manual launch, and Ari successfully obtained a remote v0.3.1 response. This verified per-user post-logon recovery, not pre-login Windows recovery.

November 2026 migration update

Following BrentPuter’s SSD failure, AriVision, AriWP, Ari-Core Filesystem and QuestNet MCP tunnels were established on Ari-Core. The QuestNet MCP server and persistent tunnel were confirmed active under systemd, with the tunnel readiness endpoint returning HTTP 200. The earlier Windows recovery tests remain historical milestones; they do not imply BrentPuter is currently operational.

Scroll to Top