/memorix-troubleshooting

name: memorix-troubleshooting
description: Use when Memorix MCP, setup, project binding, HTTP control plane, hooks, skills, or agent integration is missing, stale, or failing.

Memorix Troubleshooting

Prefer direct CLI diagnostics first. MCP is an integration surface, not the only way to operate Memorix.

Diagnostic Router

| Symptom | Check | CLI fallback | |---|---|---| | MCP tools missing | host MCP config and restart | `memorix setup --agent <agent>` | | Unsure project is bound | project status | `memorix status` and `git status` | | HTTP MCP stale or unbound | background health and explicit session bind | `memorix background status`; then `memorix session start --projectRoot <path>` | | Hooks not firing | hook status | `memorix hooks status` | | Need to inspect generated files | preview | `memorix hooks preview --agent <agent>` | | Setup looks incomplete | reinstall host integration | `memorix setup --agent <agent>` | | Need dashboard/control plane | background service | `memorix background start` or `memorix serve-http --port 3211` |

Mode Rules

• Use stdio MCP (`memorix serve`) for normal host-launched MCP integration.
• Use HTTP/background only for shared endpoint, dashboard, Docker, multiple clients, or foreground control-plane debugging.
• On Windows, prefer absolute native paths for `projectRoot` and remember that visually similar junction paths may bind different session histories.
• If MCP is unavailable but the CLI works, continue through CLI commands instead of blocking.