feat(mcp-phase2): MCP Phase 2 — Prometheus MCP + SSH MCP + alert labels

MCP-2b: prometheus_provider.py
  - prometheus_query (PromQL 即時查詢)
  - prometheus_query_range (歷史趨勢，預設 15 分鐘)
  - prometheus_get_alert_history (告警觸發歷史)
  - config: PROMETHEUS_URL + PROMETHEUS_MCP_ENABLED

MCP-2a: ssh_provider.py
  - 群組A 9 個只讀診斷工具 (top/disk/memory/logs/status/port/nginx/swap)
  - 群組B 6 個安全操作工具 (restart/compose/systemctl/clear-log/ssl/nginx-reload)
  - 四層安全守衛 (白名單/allowed_hosts/forbidden_patterns/trust_score)
  - config: SSH_MCP_ENABLED + SSH_MCP_ALLOWED_HOSTS

K8s: 04-ssh-mcp-secret.example.yaml (ssh-mcp-key Secret 範本 + 建立步驟)

Alert labels: alerts-unified.yml 補充 mcp_provider/host_type/alert_category
  覆蓋: HostHighCpuLoad/HostOutOfMemory/HostOutOfDiskSpace/DockerContainer*
        SignOzDown/SentryDown/HarborDown/GiteaDown

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/api/src/core/config.py

@@ -516,6 +516,29 @@ class Settings(BaseSettings):
             return []
         return [repo.strip() for repo in raw.split(",") if repo.strip()]
 
+    # ==========================================================================
+    # MCP Phase 2b: Prometheus MCP Server (ADR-071, 2026-04-11 Claude Sonnet 4.6)
+    # ==========================================================================
+    PROMETHEUS_URL: str = Field(
+        default="http://192.168.0.188:9090",
+        description="Prometheus server URL",
+    )
+    PROMETHEUS_MCP_ENABLED: bool = Field(
+        default=True,
+        description="啟用 Prometheus MCP Provider",
+    )
+
+    # MCP Phase 2a: SSH MCP Server (ADR-071, 2026-04-11 Claude Sonnet 4.6)
+    # ==========================================================================
+    SSH_MCP_ENABLED: bool = Field(
+        default=False,
+        description="啟用 SSH MCP Provider（需 K8s Secret ssh-mcp-key 掛載）",
+    )
+    SSH_MCP_ALLOWED_HOSTS: str = Field(
+        default="192.168.0.188,192.168.0.110,192.168.0.111",
+        description="允許 SSH 的主機 IP 清單（逗號分隔）",
+    )
+
     # ==========================================================================
     # Phase 13.2: Grafana MCP Tool (#83)
     # ==========================================================================

apps/api/src/plugins/mcp/providers/prometheus_provider.py

@@ -0,0 +1,287 @@
+"""
+Prometheus MCP Tool Provider — MCP Phase 2b
+=============================================
+提供三個 Prometheus 查詢工具，供 AI 情報收集使用：
+
+  prometheus_query           — 即時查詢（PromQL instant query）
+  prometheus_query_range     — 範圍查詢（step 預設 60s）
+  prometheus_get_alert_history — 告警觸發歷史（最近 N 筆）
+
+設計原則:
+- 唯讀（只呼叫 /api/v1/query 和 /api/v1/query_range）
+- 所有 HTTP 錯誤回傳 MCPToolResult(success=False)，不拋出
+- 逾時 10 秒，避免 AI 等待過長
+- Lazy import settings，避免 circular import
+
+建立時間: 2026-04-11 (台北時區)
+建立者: Claude Sonnet 4.6 — MCP Phase 2b
+
+@see docs/superpowers/specs/2026-04-10-infra-rebuild-sprint-abc-design.md §MCP-2b
+"""
+
+import uuid
+from datetime import UTC, datetime, timedelta
+from typing import Any
+
+import httpx
+import structlog
+
+from src.plugins.mcp.interfaces import MCPTool, MCPToolProvider, MCPToolResult
+
+logger = structlog.get_logger(__name__)
+
+_HTTP_TIMEOUT = 10.0  # 秒
+
+
+class PrometheusProvider(MCPToolProvider):
+    """
+    Prometheus MCP Tool Provider
+
+    三個工具:
+      prometheus_query          — 即時查詢
+      prometheus_query_range    — 範圍查詢（過去 N 分鐘，step 60s）
+      prometheus_get_alert_history — 告警觸發歷史
+    """
+
+    @property
+    def name(self) -> str:
+        return "prometheus"
+
+    @property
+    def enabled(self) -> bool:
+        from src.core.config import settings
+        return settings.PROMETHEUS_MCP_ENABLED
+
+    def _base_url(self) -> str:
+        from src.core.config import settings
+        return settings.PROMETHEUS_URL.rstrip("/")
+
+    async def list_tools(self) -> list[MCPTool]:
+        return [
+            MCPTool(
+                name="prometheus_query",
+                description=(
+                    "Execute a PromQL instant query against Prometheus. "
+                    "Returns current value(s) for the given expression."
+                ),
+                input_schema={
+                    "type": "object",
+                    "properties": {
+                        "query": {
+                            "type": "string",
+                            "description": "PromQL expression, e.g. 'up{job=\"awoooi-api\"}'",
+                        },
+                        "time": {
+                            "type": "string",
+                            "description": "Evaluation timestamp (RFC3339 or Unix). Default: now.",
+                        },
+                    },
+                    "required": ["query"],
+                },
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="prometheus_query_range",
+                description=(
+                    "Execute a PromQL range query. "
+                    "Returns time-series data over the last N minutes."
+                ),
+                input_schema={
+                    "type": "object",
+                    "properties": {
+                        "query": {
+                            "type": "string",
+                            "description": "PromQL expression",
+                        },
+                        "window_minutes": {
+                            "type": "integer",
+                            "description": "Look-back window in minutes (default: 15)",
+                        },
+                        "step_seconds": {
+                            "type": "integer",
+                            "description": "Step interval in seconds (default: 60)",
+                        },
+                    },
+                    "required": ["query"],
+                },
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="prometheus_get_alert_history",
+                description=(
+                    "Get recent firing history for a specific alertname. "
+                    "Returns timestamps and labels of the last N firings."
+                ),
+                input_schema={
+                    "type": "object",
+                    "properties": {
+                        "alertname": {
+                            "type": "string",
+                            "description": "Alert name (e.g. HostHighCpuLoad)",
+                        },
+                        "window_hours": {
+                            "type": "integer",
+                            "description": "Look-back window in hours (default: 24)",
+                        },
+                        "limit": {
+                            "type": "integer",
+                            "description": "Max number of results (default: 20)",
+                        },
+                    },
+                    "required": ["alertname"],
+                },
+                server_name=self.name,
+            ),
+        ]
+
+    async def execute(
+        self,
+        tool_name: str,
+        parameters: dict[str, Any],
+    ) -> MCPToolResult:
+        execution_id = str(uuid.uuid4())
+        start = datetime.now(UTC)
+
+        try:
+            if tool_name == "prometheus_query":
+                output = await self._instant_query(parameters)
+            elif tool_name == "prometheus_query_range":
+                output = await self._range_query(parameters)
+            elif tool_name == "prometheus_get_alert_history":
+                output = await self._alert_history(parameters)
+            else:
+                raise ValueError(f"Unknown tool: {tool_name}")
+
+            duration = (datetime.now(UTC) - start).total_seconds()
+            return MCPToolResult(
+                success=True,
+                execution_id=execution_id,
+                output=output,
+                duration=duration,
+            )
+
+        except Exception as e:
+            duration = (datetime.now(UTC) - start).total_seconds()
+            logger.warning(
+                "prometheus_tool_failed",
+                tool=tool_name,
+                error=str(e),
+            )
+            return MCPToolResult(
+                success=False,
+                execution_id=execution_id,
+                error=str(e),
+                duration=duration,
+            )
+
+    async def health_check(self) -> bool:
+        try:
+            async with httpx.AsyncClient(timeout=5.0) as client:
+                resp = await client.get(f"{self._base_url()}/-/ready")
+                return resp.status_code == 200
+        except Exception:
+            return False
+
+    # =========================================================================
+    # Internal helpers
+    # =========================================================================
+
+    async def _instant_query(self, params: dict) -> dict:
+        query = params["query"]
+        time_param = params.get("time", "")
+
+        url = f"{self._base_url()}/api/v1/query"
+        payload: dict[str, Any] = {"query": query}
+        if time_param:
+            payload["time"] = time_param
+
+        async with httpx.AsyncClient(timeout=_HTTP_TIMEOUT) as client:
+            resp = await client.get(url, params=payload)
+            resp.raise_for_status()
+            data = resp.json()
+
+        if data.get("status") != "success":
+            raise RuntimeError(f"Prometheus error: {data.get('error', 'unknown')}")
+
+        result = data.get("data", {}).get("result", [])
+        return {
+            "query": query,
+            "result_type": data.get("data", {}).get("resultType", ""),
+            "result": result,
+            "result_count": len(result),
+        }
+
+    async def _range_query(self, params: dict) -> dict:
+        query = params["query"]
+        window_minutes = int(params.get("window_minutes", 15))
+        step_seconds = int(params.get("step_seconds", 60))
+
+        now = datetime.now(UTC)
+        end_ts = now.timestamp()
+        start_ts = (now - timedelta(minutes=window_minutes)).timestamp()
+
+        url = f"{self._base_url()}/api/v1/query_range"
+        payload = {
+            "query": query,
+            "start": str(start_ts),
+            "end": str(end_ts),
+            "step": f"{step_seconds}s",
+        }
+
+        async with httpx.AsyncClient(timeout=_HTTP_TIMEOUT) as client:
+            resp = await client.get(url, params=payload)
+            resp.raise_for_status()
+            data = resp.json()
+
+        if data.get("status") != "success":
+            raise RuntimeError(f"Prometheus error: {data.get('error', 'unknown')}")
+
+        result = data.get("data", {}).get("result", [])
+        return {
+            "query": query,
+            "window_minutes": window_minutes,
+            "step_seconds": step_seconds,
+            "result_type": data.get("data", {}).get("resultType", ""),
+            "result": result,
+            "series_count": len(result),
+        }
+
+    async def _alert_history(self, params: dict) -> dict:
+        alertname = params["alertname"]
+        window_hours = int(params.get("window_hours", 24))
+        limit = int(params.get("limit", 20))
+
+        # 用 ALERTS_FOR_STATE metric 查詢告警觸發歷史
+        # 若找不到，fallback 用 ALERTS metric
+        query = f'ALERTS{{alertname="{alertname}",alertstate="firing"}}'
+        window_minutes = window_hours * 60
+
+        range_result = await self._range_query({
+            "query": query,
+            "window_minutes": window_minutes,
+            "step_seconds": 300,  # 5 分鐘粒度
+        })
+
+        # 整理為易讀的觸發時間列表
+        firings = []
+        for series in range_result.get("result", [])[:limit]:
+            labels = series.get("metric", {})
+            values = series.get("values", [])
+            for ts, val in values:
+                if float(val) > 0:
+                    firings.append({
+                        "timestamp": datetime.fromtimestamp(float(ts), tz=UTC).isoformat(),
+                        "labels": {k: v for k, v in labels.items() if k != "__name__"},
+                        "value": float(val),
+                    })
+
+        # 最近優先，截斷到 limit
+        firings.sort(key=lambda x: x["timestamp"], reverse=True)
+        firings = firings[:limit]
+
+        return {
+            "alertname": alertname,
+            "window_hours": window_hours,
+            "firing_count": len(firings),
+            "firings": firings,
+        }

apps/api/src/plugins/mcp/providers/ssh_provider.py

@@ -0,0 +1,495 @@
+"""
+SSH MCP Tool Provider — MCP Phase 2a
+======================================
+主機層操作代理 — 補完 K8s 無法覆蓋的 70% 主機層告警
+
+工具分組:
+  群組 A (9 個只讀診斷工具，無需信任度)
+    ssh_get_top_processes     — HostHighCpuLoad / HostOutOfMemory
+    ssh_get_disk_usage        — HostOutOfDiskSpace
+    ssh_get_memory_info       — HostOutOfMemory
+    ssh_get_container_logs    — DockerContainerExited / HarborDown
+    ssh_get_container_status  — 所有 DockerContainer* 告警
+    ssh_get_service_status    — OllamaDown / KaliScannerDown
+    ssh_check_port            — 服務宕機確認
+    ssh_get_nginx_error_log   — 網站宕機 / TLS 問題
+    ssh_get_swap_info         — HostOutOfMemory
+
+  群組 B (6 個安全操作工具，需 trust_score >= 0.8)
+    ssh_docker_restart        — DockerContainerExited / HarborDown
+    ssh_docker_compose_restart — SentryDown / SignOzDown / GiteaDown
+    ssh_systemctl_restart     — OllamaDown / KaliScannerDown
+    ssh_clear_docker_logs     — HostOutOfDiskSpace (log 佔用)
+    ssh_renew_ssl             — TLSCertExpiringIn7Days
+    ssh_reload_nginx          — TLSProbeFailure / conf 更新後
+
+四層安全守衛 (缺一不可):
+  1. tool_name 必須在白名單
+  2. host 必須在 SSH_MCP_ALLOWED_HOSTS
+  3. 所有參數通過 FORBIDDEN_PATTERNS 正則審查
+  4. 群組 B 工具必須 trust_score >= 0.8
+
+SSH 連線:
+  認證方式: Private Key，從 /run/secrets/ssh_mcp_key 讀取
+  連線庫: asyncssh (純 Python)
+  絞殺者開關: SSH_MCP_ENABLED env var
+
+建立時間: 2026-04-11 (台北時區)
+建立者: Claude Sonnet 4.6 — MCP Phase 2a
+
+@see docs/superpowers/specs/2026-04-10-infra-rebuild-sprint-abc-design.md §MCP-2a
+"""
+
+import re
+import uuid
+from datetime import UTC, datetime
+from typing import Any
+
+import structlog
+
+from src.plugins.mcp.interfaces import MCPTool, MCPToolProvider, MCPToolResult
+
+logger = structlog.get_logger(__name__)
+
+# =============================================================================
+# 安全常數
+# =============================================================================
+
+SSH_KEY_PATH = "/run/secrets/ssh_mcp_key"
+SSH_USER = "wooo"
+DIAG_TIMEOUT = 10   # 診斷類超時（秒）
+OP_TIMEOUT   = 60   # 操作類超時（秒）
+
+# 禁止字串正則（硬編碼守衛）
+FORBIDDEN_PATTERNS = [
+    r"rm\s+-rf",          # 遞歸刪除
+    r"/etc/passwd",       # 系統帳號
+    r"authorized_keys",   # SSH key
+    r"sudoers",           # 權限設定
+    r"\$\(",              # 命令替換
+    r"`",                 # 反引號執行
+    r"\|.*rm",            # pipe 到刪除
+    r">\s*/etc/",         # 重定向到系統目錄
+]
+
+# 群組 A（只讀）
+GROUP_A_TOOLS = {
+    "ssh_get_top_processes",
+    "ssh_get_disk_usage",
+    "ssh_get_memory_info",
+    "ssh_get_container_logs",
+    "ssh_get_container_status",
+    "ssh_get_service_status",
+    "ssh_check_port",
+    "ssh_get_nginx_error_log",
+    "ssh_get_swap_info",
+}
+
+# 群組 B（安全操作，需 trust_score）
+GROUP_B_TOOLS = {
+    "ssh_docker_restart",
+    "ssh_docker_compose_restart",
+    "ssh_systemctl_restart",
+    "ssh_clear_docker_logs",
+    "ssh_renew_ssl",
+    "ssh_reload_nginx",
+}
+
+ALL_TOOLS = GROUP_A_TOOLS | GROUP_B_TOOLS
+
+MIN_TRUST_SCORE_FOR_GROUP_B = 0.8
+
+
+class SSHProvider(MCPToolProvider):
+    """
+    SSH MCP Provider — 主機層操作代理
+
+    Phase MCP-2a: 補完 K8s 無法覆蓋的 70% 主機層告警
+    絞殺者開關: SSH_MCP_ENABLED env var
+    """
+
+    @property
+    def name(self) -> str:
+        return "ssh_host"
+
+    @property
+    def enabled(self) -> bool:
+        from src.core.config import settings
+        return getattr(settings, "SSH_MCP_ENABLED", False)
+
+    def _allowed_hosts(self) -> list[str]:
+        from src.core.config import settings
+        raw = getattr(settings, "SSH_MCP_ALLOWED_HOSTS", "")
+        if not raw:
+            return ["192.168.0.188", "192.168.0.110", "192.168.0.111"]
+        if isinstance(raw, list):
+            return raw
+        return [h.strip() for h in raw.split(",") if h.strip()]
+
+    # =========================================================================
+    # list_tools
+    # =========================================================================
+
+    async def list_tools(self) -> list[MCPTool]:
+        return [
+            # ---- 群組 A ----
+            MCPTool(
+                name="ssh_get_top_processes",
+                description="Get top CPU/memory consuming processes on the target host (ps aux --sort=-%cpu | head 15). Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string", "description": "Target host IP (e.g. 192.168.0.188)"},
+                }, "required": ["host"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_disk_usage",
+                description="Get disk usage on the target host (df -h && du -sh /var/lib/docker). Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                }, "required": ["host"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_memory_info",
+                description="Get memory info (free -h). Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                }, "required": ["host"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_container_logs",
+                description="Get Docker container logs (last 50 lines). Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "container_name": {"type": "string", "description": "Docker container name"},
+                    "tail": {"type": "integer", "description": "Number of lines (default: 50)"},
+                }, "required": ["host", "container_name"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_container_status",
+                description="Get Docker container status filtered by name. Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "filter_name": {"type": "string", "description": "Container name filter"},
+                }, "required": ["host", "filter_name"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_service_status",
+                description="Get systemd service status. Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "service": {"type": "string", "description": "Service name (e.g. ollama)"},
+                }, "required": ["host", "service"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_check_port",
+                description="Check if a port is listening on the target host. Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "port": {"type": "integer", "description": "Port number"},
+                }, "required": ["host", "port"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_nginx_error_log",
+                description="Get last 50 lines of nginx error log. Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "lines": {"type": "integer", "description": "Number of lines (default: 50)"},
+                }, "required": ["host"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_get_swap_info",
+                description="Get swap and memory info (swapon --show && free -h). Read-only.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                }, "required": ["host"]},
+                server_name=self.name,
+            ),
+            # ---- 群組 B ----
+            MCPTool(
+                name="ssh_docker_restart",
+                description="Restart a Docker container (docker restart <name>). Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "container_name": {"type": "string"},
+                    "trust_score": {"type": "number", "description": "Current trust score (0.0-1.0)"},
+                }, "required": ["host", "container_name", "trust_score"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_docker_compose_restart",
+                description="Restart a service via docker compose. Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "compose_dir": {"type": "string", "description": "Directory containing docker-compose.yml"},
+                    "service": {"type": "string", "description": "Service name in compose file"},
+                    "trust_score": {"type": "number"},
+                }, "required": ["host", "compose_dir", "service", "trust_score"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_systemctl_restart",
+                description="Restart a systemd service (systemctl restart <service>). Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "service": {"type": "string"},
+                    "trust_score": {"type": "number"},
+                }, "required": ["host", "service", "trust_score"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_clear_docker_logs",
+                description="Truncate Docker container log file to free disk space. Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "container_name": {"type": "string"},
+                    "trust_score": {"type": "number"},
+                }, "required": ["host", "container_name", "trust_score"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_renew_ssl",
+                description="Renew SSL certificate via certbot. Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "domain": {"type": "string", "description": "Certificate name (certbot --cert-name)"},
+                    "trust_score": {"type": "number"},
+                }, "required": ["host", "domain", "trust_score"]},
+                server_name=self.name,
+            ),
+            MCPTool(
+                name="ssh_reload_nginx",
+                description="Test and reload nginx config (nginx -t && systemctl reload nginx). Requires trust_score >= 0.8.",
+                input_schema={"type": "object", "properties": {
+                    "host": {"type": "string"},
+                    "trust_score": {"type": "number"},
+                }, "required": ["host", "trust_score"]},
+                server_name=self.name,
+            ),
+        ]
+
+    # =========================================================================
+    # execute
+    # =========================================================================
+
+    async def execute(
+        self,
+        tool_name: str,
+        parameters: dict[str, Any],
+    ) -> MCPToolResult:
+        execution_id = str(uuid.uuid4())
+        start = datetime.now(UTC)
+
+        # 守衛 1: 白名單
+        if tool_name not in ALL_TOOLS:
+            return MCPToolResult(
+                success=False,
+                execution_id=execution_id,
+                error=f"Unknown tool: {tool_name}",
+            )
+
+        host = parameters.get("host", "")
+
+        # 守衛 2: 允許的 host
+        if host not in self._allowed_hosts():
+            return MCPToolResult(
+                success=False,
+                execution_id=execution_id,
+                error=f"Host '{host}' not in SSH_MCP_ALLOWED_HOSTS",
+            )
+
+        # 守衛 3: 參數安全檢查
+        security_error = self._check_params_safety(parameters)
+        if security_error:
+            logger.warning(
+                "ssh_mcp_forbidden_pattern",
+                tool=tool_name,
+                host=host,
+                reason=security_error,
+            )
+            return MCPToolResult(
+                success=False,
+                execution_id=execution_id,
+                error=f"Security guard blocked: {security_error}",
+            )
+
+        # 守衛 4: 群組 B 信任度
+        if tool_name in GROUP_B_TOOLS:
+            trust_score = float(parameters.get("trust_score", 0.0))
+            if trust_score < MIN_TRUST_SCORE_FOR_GROUP_B:
+                return MCPToolResult(
+                    success=False,
+                    execution_id=execution_id,
+                    error=(
+                        f"Group B tool '{tool_name}' requires trust_score >= "
+                        f"{MIN_TRUST_SCORE_FOR_GROUP_B}, got {trust_score:.2f}. "
+                        "Consider manual execution."
+                    ),
+                )
+
+        # 執行
+        try:
+            is_group_b = tool_name in GROUP_B_TOOLS
+            timeout = OP_TIMEOUT if is_group_b else DIAG_TIMEOUT
+            output = await self._run_tool(tool_name, parameters, host, timeout)
+            duration = (datetime.now(UTC) - start).total_seconds()
+            logger.info("ssh_mcp_executed", tool=tool_name, host=host, duration=duration)
+            return MCPToolResult(
+                success=True,
+                execution_id=execution_id,
+                output=output,
+                duration=duration,
+            )
+        except Exception as e:
+            duration = (datetime.now(UTC) - start).total_seconds()
+            logger.warning("ssh_mcp_failed", tool=tool_name, host=host, error=str(e))
+            return MCPToolResult(
+                success=False,
+                execution_id=execution_id,
+                error=str(e),
+                duration=duration,
+            )
+
+    async def health_check(self) -> bool:
+        """只確認 SSH key 存在，不真正 SSH（避免 health check 觸發副作用）"""
+        import os
+        return os.path.exists(SSH_KEY_PATH)
+
+    # =========================================================================
+    # 安全守衛
+    # =========================================================================
+
+    def _check_params_safety(self, params: dict) -> str | None:
+        """
+        掃描所有字串參數，找到禁止字串即返回 error 字串
+        """
+        for value in params.values():
+            if not isinstance(value, str):
+                continue
+            for pattern in FORBIDDEN_PATTERNS:
+                if re.search(pattern, value, re.IGNORECASE):
+                    return f"Forbidden pattern '{pattern}' found in parameter"
+        return None
+
+    # =========================================================================
+    # 工具執行
+    # =========================================================================
+
+    async def _run_tool(
+        self,
+        tool_name: str,
+        params: dict,
+        host: str,
+        timeout: int,
+    ) -> dict:
+        cmd = self._build_command(tool_name, params)
+        stdout, stderr = await self._ssh_exec(host, cmd, timeout)
+        return {
+            "host": host,
+            "tool": tool_name,
+            "command": cmd,
+            "stdout": stdout,
+            "stderr": stderr,
+        }
+
+    def _build_command(self, tool_name: str, params: dict) -> str:
+        if tool_name == "ssh_get_top_processes":
+            return "ps aux --sort=-%cpu | head -15"
+
+        if tool_name == "ssh_get_disk_usage":
+            return "df -h && echo '---' && du -sh /var/lib/docker 2>/dev/null || true"
+
+        if tool_name == "ssh_get_memory_info":
+            return "free -h"
+
+        if tool_name == "ssh_get_container_logs":
+            name = params["container_name"]
+            tail = int(params.get("tail", 50))
+            return f"docker logs {name} --tail {tail} 2>&1"
+
+        if tool_name == "ssh_get_container_status":
+            name = params["filter_name"]
+            return f"docker ps -a --filter name={name}"
+
+        if tool_name == "ssh_get_service_status":
+            svc = params["service"]
+            return f"systemctl status {svc} --no-pager -l 2>&1 | head -30"
+
+        if tool_name == "ssh_check_port":
+            port = int(params["port"])
+            return f"ss -tlnp | grep :{port}"
+
+        if tool_name == "ssh_get_nginx_error_log":
+            lines = int(params.get("lines", 50))
+            return f"tail -n {lines} /var/log/nginx/error.log 2>/dev/null || echo 'Log not found'"
+
+        if tool_name == "ssh_get_swap_info":
+            return "swapon --show; echo '---'; free -h"
+
+        if tool_name == "ssh_docker_restart":
+            name = params["container_name"]
+            return f"docker restart {name}"
+
+        if tool_name == "ssh_docker_compose_restart":
+            compose_dir = params["compose_dir"]
+            service = params["service"]
+            return f"cd {compose_dir} && docker compose restart {service}"
+
+        if tool_name == "ssh_systemctl_restart":
+            svc = params["service"]
+            return f"systemctl restart {svc}"
+
+        if tool_name == "ssh_clear_docker_logs":
+            name = params["container_name"]
+            # 透過 docker inspect 取得 log 路徑，再截斷
+            return (
+                f"LOG_PATH=$(docker inspect --format='{{{{.LogPath}}}}' {name} 2>/dev/null) "
+                f"&& [ -n \"$LOG_PATH\" ] && truncate -s 0 \"$LOG_PATH\" && echo 'Cleared' "
+                f"|| echo 'Container not found'"
+            )
+
+        if tool_name == "ssh_renew_ssl":
+            domain = params["domain"]
+            return f"/snap/bin/certbot renew --cert-name {domain} --non-interactive 2>&1"
+
+        if tool_name == "ssh_reload_nginx":
+            return "nginx -t 2>&1 && systemctl reload nginx && echo 'Nginx reloaded'"
+
+        raise ValueError(f"No command builder for tool: {tool_name}")
+
+    async def _ssh_exec(self, host: str, cmd: str, timeout: int) -> tuple[str, str]:
+        """
+        透過 asyncssh 執行 SSH 命令
+
+        如果 asyncssh 未安裝，返回明確錯誤而非崩潰。
+        """
+        try:
+            import asyncssh
+        except ImportError:
+            raise RuntimeError(
+                "asyncssh is not installed. "
+                "Add 'asyncssh' to pyproject.toml dependencies."
+            )
+
+        import os
+        if not os.path.exists(SSH_KEY_PATH):
+            raise RuntimeError(
+                f"SSH key not found at {SSH_KEY_PATH}. "
+                "Ensure K8s Secret 'ssh-mcp-key' is mounted correctly."
+            )
+
+        async with asyncssh.connect(
+            host,
+            username=SSH_USER,
+            client_keys=[SSH_KEY_PATH],
+            known_hosts=None,  # 內網信任，不驗證 known_hosts
+            connect_timeout=timeout,
+        ) as conn:
+            result = await asyncssh.run(conn, cmd, timeout=timeout, check=False)
+            return (result.stdout or ""), (result.stderr or "")

k8s/awoooi-prod/04-ssh-mcp-secret.example.yaml

@@ -0,0 +1,51 @@
+# SSH MCP Key Secret — MCP Phase 2a (ADR-071)
+# ============================================
+# 此 Secret 供 SSH MCP Provider 使用，讓 API Pod 可 SSH 到 188/110/111 主機執行診斷和修復指令。
+#
+# 建立步驟:
+#   1. 在 API Pod 宿主機生成 SSH key pair:
+#      ssh-keygen -t ed25519 -f /tmp/ssh-mcp-key -N "" -C "awoooi-mcp@k3s"
+#
+#   2. 將公鑰加入允許主機的 authorized_keys:
+#      ssh-copy-id -i /tmp/ssh-mcp-key.pub wooo@192.168.0.188
+#      ssh-copy-id -i /tmp/ssh-mcp-key.pub wooo@192.168.0.110
+#      ssh-copy-id -i /tmp/ssh-mcp-key.pub wooo@192.168.0.111
+#
+#   3. 建立 K8s Secret:
+#      kubectl create secret generic ssh-mcp-key \
+#        --from-file=ssh_mcp_key=/tmp/ssh-mcp-key \
+#        -n awoooi-prod
+#
+#   4. 刪除暫存 key:
+#      rm /tmp/ssh-mcp-key /tmp/ssh-mcp-key.pub
+#
+#   5. 在 K8s Deployment 加入 volume mount (參考 apps/api/k8s-deployment.yaml)
+#      volumeMounts:
+#        - name: ssh-mcp-key
+#          mountPath: /run/secrets/ssh_mcp_key
+#          subPath: ssh_mcp_key
+#          readOnly: true
+#      volumes:
+#        - name: ssh-mcp-key
+#          secret:
+#            secretName: ssh-mcp-key
+#            defaultMode: 0400
+#
+#   6. 設定環境變數 SSH_MCP_ENABLED=true（在 K8s ConfigMap 或 Deployment env 加入）
+#
+# ⚠️ 此檔案是範本，實際 Secret 由 CI/CD 注入，禁止提交私鑰到 Git
+# ⚠️ 04-ssh-mcp-secret.yaml 已加入 .gitignore
+
+apiVersion: v1
+kind: Secret
+metadata:
+  name: ssh-mcp-key
+  namespace: awoooi-prod
+  annotations:
+    # MCP Phase 2a (ADR-071, 2026-04-11)
+    description: "SSH private key for MCP SSH Provider — allows API pods to diagnose/repair host-layer services"
+type: Opaque
+data:
+  # Base64 encoded private key
+  # 使用方式: echo "$(cat /tmp/ssh-mcp-key | base64 -w 0)"
+  ssh_mcp_key: "CHANGE_ME_BASE64_ENCODED_PRIVATE_KEY"

ops/monitoring/alerts-unified.yml

@@ -39,6 +39,10 @@ groups:
           layer: systemd-188
           team: ops
           auto_repair: "true"
+          # MCP Phase 2a (ADR-071, 2026-04-11 Claude Sonnet 4.6): SSH MCP 路由標籤
+          mcp_provider: "ssh_host"
+          host_type: "bare_metal"
+          alert_category: "host_resource"
         annotations:
           summary: "主機 {{ $labels.host }} CPU 高負載"
           description: "CPU 使用率超過 80%"
@@ -51,6 +55,9 @@ groups:
           layer: systemd-188
           team: ops
           auto_repair: "false"
+          mcp_provider: "ssh_host"
+          host_type: "bare_metal"
+          alert_category: "host_resource"
         annotations:
           summary: "主機 {{ $labels.host }} 記憶體不足"
           description: "記憶體使用率超過 85%"
@@ -63,6 +70,9 @@ groups:
           layer: systemd-188
           team: ops
           auto_repair: "false"
+          mcp_provider: "ssh_host"
+          host_type: "bare_metal"
+          alert_category: "host_resource"
         annotations:
           summary: "主機 {{ $labels.host }} 磁碟空間不足"
           description: "磁碟使用率超過 85%"
@@ -329,6 +339,9 @@ groups:
           host: "188"
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          target_host: "192.168.0.188"
+          alert_category: "devops_tool"
         annotations:
           summary: "SignOz 服務離線"
           description: "SignOz (192.168.0.188:3301) 已離線超過 2 分鐘"
@@ -344,6 +357,9 @@ groups:
           host: "110"
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          target_host: "192.168.0.110"
+          alert_category: "devops_tool"
         annotations:
           summary: "Sentry 服務離線"
           description: "Sentry (192.168.0.110:9000) 已離線超過 2 分鐘"
@@ -358,6 +374,9 @@ groups:
           host: "110"
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          target_host: "192.168.0.110"
+          alert_category: "devops_tool"
         annotations:
           summary: "Harbor Registry 離線"
           description: "Harbor (192.168.0.110:5000) 已離線超過 2 分鐘，CD pipeline 將無法拉取映像"
@@ -372,6 +391,9 @@ groups:
           host: "110"
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          target_host: "192.168.0.110"
+          alert_category: "devops_tool"
         annotations:
           summary: "Gitea Git 服務離線"
           description: "Gitea (192.168.0.110:3001) 已離線超過 2 分鐘，CD pipeline 失效"
@@ -494,6 +516,8 @@ groups:
           layer: docker
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          alert_category: "devops_tool"
         annotations:
           summary: "容器 {{ $labels.container }} 健康檢查失敗"
           description: "主機 {{ $labels.host }} 容器 {{ $labels.container }} 健康狀態異常，持續 2 分鐘"
@@ -506,6 +530,8 @@ groups:
           layer: docker
           team: ops
           auto_repair: "true"
+          mcp_provider: "ssh_host"
+          alert_category: "devops_tool"
         annotations:
           summary: "容器 {{ $labels.container }} 已停止"
           description: "主機 {{ $labels.host }} 容器 {{ $labels.container }} 已 exited/dead，持續 1 分鐘"