Initial commit: Agentic OS troubleshooting platform

Self-hosted, Docker-based agentic troubleshooting platform: FastAPI backend +
LangGraph agent, Next.js UI, tiered LLM routing (local Ollama -> Gemini ->
DeepSeek -> OpenRouter), MCP server manager, encrypted device credentials,
RBAC, audit log, project-memory + Obsidian integrations, and editable
troubleshooting decision rules tuned for the GeneseasX vessel stack.

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/app/mcp_manager/recovery.py

@@ -0,0 +1,46 @@
+"""Auto-reload MCP servers when a tool call fails due to server unavailability."""
+from __future__ import annotations
+
+import logging
+from collections.abc import Awaitable, Callable
+
+logger = logging.getLogger(__name__)
+
+EmitFn = Callable[[int, str, str, dict | None], Awaitable[None]]
+
+
+async def call_tool_resilient(
+    manager,
+    server: str,
+    tool: str,
+    args: dict,
+    *,
+    task_id: int | None = None,
+    emit: EmitFn | None = None,
+    max_retries: int = 1,
+) -> dict:
+    """Call MCP tool; on server-unavailable errors, reload that server and retry once."""
+    last_exc: Exception | None = None
+    for attempt in range(max_retries + 1):
+        try:
+            return await manager.call_tool(server, tool, args)
+        except RuntimeError as exc:
+            last_exc = exc
+            msg = str(exc).lower()
+            recoverable = "not available" in msg or "not loaded" in msg
+            if not recoverable or attempt >= max_retries:
+                raise
+            logger.warning("MCP %s unavailable for %s — reloading (attempt %s)", server, tool, attempt + 1)
+            if task_id and emit:
+                await emit(
+                    task_id,
+                    "log",
+                    f"MCP '{server}' unavailable — reloading server and retrying {tool}…",
+                    {"server": server, "tool": tool},
+                )
+            try:
+                await manager.reload_server(server)
+            except Exception as reload_exc:  # noqa: BLE001
+                logger.warning("MCP reload %s failed: %s", server, reload_exc)
+                raise exc from reload_exc
+    raise last_exc or RuntimeError(f"MCP call failed: {server}/{tool}")