feat(hosts): the hottest sensor is named with its chip, and the host-reading cutoffs are read and set over MCP (#6806)

Agent 1.27.0 sends each temperature sensor's hwmon chip. Steward keeps the
chip per host and label beside the reading, so a sensor's metric key stays
its label, and the strip and the heat alert name it as "NVMe · Sensor 2"
rather than a bare "Sensor 2".

get_thresholds (viewer) and set_thresholds (admin, audited) read and change
the warn/crit cutoffs CPU, memory, disk, load and temperature are graded by,
as Settings → Thresholds does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

steward/mcp/tools/__init__.py

@@ -13,10 +13,10 @@ each argument's empty value means, and what comes back.
 from __future__ import annotations
 
 from . import (alerts, anomalies, audit, automation, backups, hosts, metrics, monitors, overview, session,
-               tickets, timeline)
+               thresholds, tickets, timeline)
 
-_MODULES = (session, overview, hosts, monitors, alerts, anomalies, metrics, audit, timeline, tickets,
-            automation, backups)
+_MODULES = (session, overview, hosts, monitors, alerts, anomalies, thresholds, metrics, audit, timeline,
+            tickets, automation, backups)
 
 
 def register_all(register_tool) -> None:

steward/mcp/tools/thresholds.py

@@ -0,0 +1,94 @@
+"""The cutoffs a host's readings are graded by, as Settings → Thresholds sets
+them: amber past `warn`, red past `crit`."""
+from __future__ import annotations
+
+from steward.core.settings import DEFAULTS, get_all_settings, set_setting, to_thresholds_cfg
+from steward.mcp._context import mcp_app
+from steward.models.users import UserRole
+
+from ._audit import audit_mcp
+from ._shape import db_session
+
+# Each graded reading, its unit, and what it measures.
+READINGS = {
+    "cpu": ("%", "CPU use"),
+    "mem": ("%", "memory used"),
+    "disk": ("%", "fullest mount's space used"),
+    "load": ("%", "1-minute load per core"),
+    "temp": ("°C", "hottest real thermometer"),
+}
+
+
+def _view(settings: dict) -> dict:
+    cfg = to_thresholds_cfg(settings)
+    return {kind: {"warn": cfg[kind]["warn"], "crit": cfg[kind]["crit"], "unit": unit, "measures": what,
+                   "default": {"warn": DEFAULTS[f"thresholds.{kind}_warn"],
+                               "crit": DEFAULTS[f"thresholds.{kind}_crit"]}}
+            for kind, (unit, what) in READINGS.items()}
+
+
+async def get_thresholds() -> dict:
+    """The cutoffs host readings are graded by, as Settings → Thresholds
+    sets them: per reading (cpu, mem, disk, load, temp), `warn` (amber past
+    it) and `crit` (red past it) in `unit`, with Steward's default for each.
+    The host pages, the Overview and get_overview grade by these. Alert
+    rules have their own cutoffs (list_alert_rules); these don't fire
+    anything. set_thresholds (admin) changes them.
+    """
+    async with db_session() as db:
+        return _view(await get_all_settings(db))
+
+
+async def set_thresholds(cutoffs: dict[str, dict[str, float]]) -> dict:
+    """Change the cutoffs host readings are graded by, as Settings →
+    Thresholds does. Takes effect on the next page load; audited.
+
+    cutoffs: {reading: {"warn": n, "crit": n}} for any of get_thresholds'
+    readings, e.g. {"temp": {"crit": 90}}. Either of warn and crit may be
+    left out and keeps its value; warn must stay below crit. A reading not
+    named is left as it is. Returns the cutoffs as they now stand.
+    """
+    unknown = sorted(set(cutoffs or {}) - set(READINGS))
+    if unknown:
+        raise ValueError(f"No reading named {', '.join(unknown)}; they are {', '.join(READINGS)}.")
+    if not cutoffs:
+        raise ValueError("Nothing to change: pass cutoffs, e.g. {\"temp\": {\"crit\": 90}}.")
+    async with db_session() as db:
+        current = _view(await get_all_settings(db))
+    changes: dict = {}
+    for kind, given in cutoffs.items():
+        extra = sorted(set(given or {}) - {"warn", "crit"})
+        if extra or not given:
+            raise ValueError(f"{kind} takes warn and/or crit, not {', '.join(extra) or 'nothing'}.")
+        new = {}
+        for edge in ("warn", "crit"):
+            if edge not in given:
+                new[edge] = current[kind][edge]
+                continue
+            try:
+                value = float(given[edge])
+            except (TypeError, ValueError):
+                raise ValueError(f"{kind} {edge} must be a number.") from None
+            if not value > 0:
+                raise ValueError(f"{kind} {edge} must be more than 0.")
+            new[edge] = int(value) if value.is_integer() else value
+            changes[f"thresholds.{kind}_{edge}"] = new[edge]
+        if new["warn"] >= new["crit"]:
+            raise ValueError(f"{kind} warn ({new['warn']}) must be below its crit ({new['crit']}).")
+    app = mcp_app()
+    async with db_session() as db:
+        async with db.begin():
+            before = await get_all_settings(db)
+            for key, value in changes.items():
+                await set_setting(db, key, value)
+        fresh = await get_all_settings(db)
+    app.config["THRESHOLDS"] = to_thresholds_cfg(fresh)
+    await audit_mcp("settings.saved", "settings", "thresholds",
+                    {"section": "thresholds",
+                     "changes": {k: {"from": before.get(k), "to": v} for k, v in changes.items()}})
+    return _view(fresh)
+
+
+def register(register_tool) -> None:
+    register_tool(get_thresholds, role=UserRole.viewer, read_only=True)
+    register_tool(set_thresholds, role=UserRole.admin, read_only=False)