AI导哈

监控集成服务

一个将AI助手与Grafana、Prometheus、Loki等监控系统集成的服务器，支持查询指标、搜索日志、管理仪表板、静默警报和关联事件。

  • by sandraschi
  • Python
  • 本地部署
  • 2 星标
  • 更新于
在 GitHub 查看 sandraschi/monitoring-mcp

使用说明

monitoring-mcp

An MCP server that connects AI assistants to your Grafana + Prometheus + Loki (+ Alertmanager) observability stack — local, Docker, Podman, or remote. Query metrics, search logs, manage dashboards, silence alerts, and correlate incidents through natural language.

Docs: INSTALL.md · Configuration · Tools · Podman · Development · Troubleshooting

Features

  • Grafana — dashboards, panels, folders, datasources, unified alerting, export/import
  • Prometheus — PromQL (instant + range), targets, rules, flags/config, sampling + cache
  • Alertmanager — list/create/expire silences
  • Loki — LogQL, WebSocket tail (with fallback), export, compare, ruler alerts
  • Cross-system correlation — incident / root-cause / bottleneck / dependency heuristics
  • Health monitoring — connectivity, storage, security posture, capacity signals
  • AI chat — local LLM via Ollama / LM Studio in the web UI

Quick Install

git clone https://github.com/sandraschi/monitoring-mcp
cd monitoring-mcp
just

Or drag the .mcpb bundle into Claude Desktop. See INSTALL.md for all options.

What You Can Do

You say The server does
"Check system health" Hits Grafana/Prometheus/Loki APIs; returns status per component
"Show my top 5 error log sources" LogQL query across Loki; groups by label
"List Grafana dashboards tagged prod" Searches Grafana API; filters by tag
"Correlate the 5xx spike at 14:30" Cross-references Prometheus error rates with Loki error logs
"Silence HighError for 2 hours" Creates an Alertmanager silence

Quick Start (Code)

await monitoring_status(operation="system_health")
await prometheus_monitoring(operation="query_metrics", query="up")
await loki_logging(operation="query_logs", query='{job="api"} |= "ERROR"')
await grafana_management(operation="list_dashboards")
await cross_system_correlation(operation="correlate_incident", incident_description="5xx spike")

Containers

# MCP only (talks to host Grafana/Prom/Loki via host gateway)
docker compose up -d --build
# or: podman compose -f docker-compose.yml up -d --build

# Optional full stack
docker compose --profile stack up -d

See docs/PODMAN.md for Podman host URLs (host.containers.internal).

Project Structure

├── src/monitoring_mcp/       # Python MCP server
│   ├── tools/                # Portmanteau tools (grafana, prometheus, loki, correlation, status)
│   ├── config.py             # Pydantic v2 settings
│   ├── utils.py              # Cache, sampling, encryption helpers
│   ├── server.py             # ASGI entry point
│   └── web.py                # REST endpoints (/api/chat, /api/health, /api/tools)
├── web_sota/                 # React/Vite/Tailwind dashboard
├── native/                   # Tauri 2.0 NSIS wrapper
├── docker-compose.yml        # Docker / Podman Compose
└── docs/                     # Reference documentation

Requirements

  • Python 3.12+ (3.13 supported)
  • FastMCP 3.4.4+
  • Optional: Grafana, Prometheus, Loki, Alertmanager reachable over HTTP

相关 MCP