Best Monitoring MCP Servers for AI agents.

Netdata anchors the category — at 79,000+ stars it's among the most-starred projects with an MCP integration anywhere, offering real-time infrastructure monitoring with metrics, logs, alerts, and ML-based anomaly detection. sentry-mcp brings error monitoring, issue tracking, and debugging context from Sentry; Dynatrace-mcp exposes logs, metrics, problems, and vulnerabilities through DQL and Davis AI; and SigNoz MCP Server covers metrics, traces, logs, alerts, and dashboards for the open-source SigNoz stack.

Between them, most observability setups are reachable — the choice usually follows whichever platform you already run.

A pattern worth copying runs through this category: several servers are deliberately read-only. lastsaas describes itself as a read-only admin server for dashboards, users, tenants, billing, logs, and health; HomeLab Monitor ships a built-in read-only MCP server over hosts, Docker, GPU, and services. MikroMCP goes further for write operations on MikroTik routers — typed tools, dry-run mode, RBAC, audit logs, and rollback.

That restraint isn't timidity. An agent that can read everything and change nothing is immediately useful and almost impossible to make dangerous, which is exactly where monitoring integrations should start.

Beyond the platforms, focused servers fill specific gaps. lighthouse-mcp runs Google Lighthouse performance audits as tool calls, useful for agents that ship frontend changes and need to verify they didn't regress page speed. The docker and pare-docker servers return container operations — ps, images, logs, build — as typed JSON, and monitor provides Valkey observability for Redis-compatible deployments.

Match the server to your existing stack first — a Dynatrace bridge is worthless without Dynatrace. Then evaluate query power: servers exposing a real query language (like DQL through Dynatrace-mcp) let agents ask precise questions instead of paging through dashboards. Watch data volume hardest of all: observability data is the most context-hungry input an agent can ingest, so servers that aggregate, filter, or summarize server-side are worth a tier of preference. And confirm write scope — alert acknowledgment and silencing are mutations, even in 'monitoring.'

Everything listed self-hosts for free as open source. The hosted variant has clean per-call economics: a maintainer running a monitoring endpoint — anomaly checks, Lighthouse audits, log analysis — can claim their Loomal listing and price each query from $0.01 in USDC via x402, with the agent paying automatically before the handler runs and settlement on Base in about two seconds. Loomal's 5% fee on settled transactions is currently waived.