Monitoring AIops

Disclaimer: Community-maintained open-source project. Not affiliated with, endorsed by, or sponsored by SolarWinds, Paessler, Zabbix, or any monitoring vendor. SolarWinds, Orion, SWQL, THWACK, PRTG, Paessler and Zabbix are trademarks of their respective owners. MIT licensed.

Governed AI-ops for network / infrastructure monitoring across three NOC platforms in one server — SolarWinds Orion (SWIS REST + SWQL, port 17778, HTTP Basic auth), Paessler PRTG (web API, port 443/8080, API token), and Zabbix 6.x/7.x (JSON-RPC 2.0 at /api_jsonrpc.php, API token) — with a built-in governance harness: unified audit log, policy engine, token/runaway budget guard, undo-token recording, and graduated-autonomy risk tiers. One config can span all NOCs; each target names its own platform.

What it does

Answers the questions a NOC operator actually repeats, and guards the writes that follow:

• Canned-SWQL library — the most-asked THWACK questions shipped as named, validated queries (nodes_down, flapping_interfaces, muted_report, high_cpu_nodes, volumes_full, unmanaged_scheduled), plus a validated read-only SWQL passthrough (SELECT-only) for everything else.
• Active-alert dedup / rollup — collapses an interface-flap or node-down storm into a single counted entry instead of a wall of near-identical alerts, across SolarWinds, PRTG, and Zabbix (problems, with the 0-5 severity scale mapped to canonical levels).
• Triple SolarWinds + PRTG + Zabbix coverage — one MCP server spans all three NOCs; no incumbent hobby MCP does.
• Governed writes — mute/unmute, maintenance windows (incl. Zabbix maintenance.create with a replayable delete-undo), unmanage/remanage, node removal, and PRTG sensor pause/resume — each audited, risk-tiered, and the destructive ones gated with dry-run + double-confirm. Suppression and maintenance writes are time-boxed (they require an end time / duration).

Security: read-only mode

This tool is meant to be handed to an AI agent, so its safety story is enforced by the server rather than requested in a prompt:

export MONITORING_READ_ONLY=1

With that set, the 13 write tools are never registered. An MCP client lists 29 tools instead of 42 — the writes are not hidden, not gated behind a flag, and not merely refused when called. They are absent from the session. A model cannot invoke a tool it was never offered, and cannot be argued into one.

That distinction is the whole point. A tool that exists but refuses still invites retry loops and "I'll describe the call instead" behaviour from smaller models, and it leaves a reviewer trusting a promise. An absent tool is a fact you can check: connect, list the tools, and see that the writes are not there.

Enforcement is two layers deep, so the switch cannot be sidestepped by changing entry point:

Read operations are unaffected, and every call is still audited to ~/.monitoring-aiops/audit.db.

The read/write split is derived from each tool's declared risk_level, and a test asserts that this never disagrees with the [READ]/[WRITE] tag in the tool's own documentation — so a write can't quietly present itself as a read.

Running a smaller / local model? See agent-guardrails.md — it lists the guardrails this tool now enforces for you (so you don't spend prompt budget restating them) and gives a ready-made system prompt for what's left.

Capability matrix (42 MCP tools)

The CLI exposes a convenience subset; the full 42-tool surface is via the MCP server.

Quick start

uv tool install monitoring-aiops          # or: pipx install monitoring-aiops
monitoring-aiops init                     # wizard: pick platform (solarwinds/prtg/zabbix) + store the secret (encrypted)
monitoring-aiops doctor                   # verify config, secrets, connectivity
monitoring-aiops overview                 # NOC summary: platform + active/unacked alerts + top rollup
monitoring-aiops swql library             # list the canned SWQL queries
monitoring-aiops swql canned nodes_down   # run a named canned query

Run as an MCP server (stdio):

export MONITORING_AIOPS_MASTER_PASSWORD=...   # unlock secrets non-interactively
monitoring-aiops mcp

Governance

Every MCP tool passes through the bundled @governed_tool harness:

• Audit — every call (params, result, status, duration, risk tier, approver, rationale) is logged to ~/.monitoring-aiops/audit.db (relocatable via MONITORING_AIOPS_HOME).
• Budget / runaway guard — token and call budgets trip a circuit breaker on tight poll/retry loops.
• Risk tiers — graduated autonomy; high-risk ops (unmanage_node, remove_node, zabbix_delete_maintenance) can require a named approver (MONITORING_AUDIT_APPROVED_BY / MONITORING_AUDIT_RATIONALE).
• Undo recording — reversible writes record an inverse descriptor (mute_alerts→unmute, unmanage_node→remanage, pause_sensor→resume, zabbix_create_maintenance→delete that maintenance id).

Supported scope & limitations

• Platforms: SolarWinds Orion (SWIS REST + SWQL), Paessler PRTG (web API), and Zabbix 6.x/7.x (JSON-RPC 2.0; API token via Bearer header on 6.4+/7.x, with a legacy auth-field fallback for 6.0).
• Zabbix scope: problems/triggers/hosts/host groups/events, bounded item history, maintenance windows (create/delete), and event acknowledge via the cross-platform alert_acknowledge. Template/discovery/user CRUD and trend.get are not covered — open an issue if you need them.
• Validation status. Behaviour is exercised by the test suite against mocked SWIS/PRTG/Zabbix responses; it has not been run against a live NOC (see docs/VERIFICATION.md). PRTG has a free perpetual 100-sensor Freeware edition with the API, and Zabbix is fully open source (a Docker-compose appliance is a 10-minute live check) — the easiest live checks. SolarWinds is a 30-day trial only; past the trial this tool is mock-only, which is the largest verification debt. monitoring-aiops doctor is the fastest live check (a SWQL query for SolarWinds, /api/status.json for PRTG, unauthenticated apiinfo.version + an authed host count for Zabbix).

Missing a capability?

Want another canned query, a platform dialect fixed, or a capability that isn't here? Open an issue or a PR — feedback and contributions are welcome.