observability-aiops
This server lets you inspect, analyze, and safely operate a self-hosted Prometheus/Alertmanager/Grafana/Loki observability stack with audited, dry-runnable writes and undo.
Query Prometheus with PromQL instant/range queries, label values, series metadata, targets, rules, alerts, config/TSDB status, and health overviews.
Review Alertmanager alerts and silences, and run RCA/noise/flap/scrape-health analyses.
Inspect Grafana dashboards, datasources, datasource health, and folders.
Read and analyze Loki logs with bounded LogQL queries, error tails, error-burst RCA, volume/cardinality analysis, and alert-to-log correlation.
Perform governed writes: create/expire silences, create annotations, update/delete dashboards, and reload Prometheus config.
Preview writes with dry_run, record undo tokens for reversible changes, list/apply undo, and audit every call with risk tiers.
Enables interacting with Grafana to list, fetch, update, and delete dashboards, manage datasources and folders, check datasource health, and create annotations, all with audit logging and undo support.
Provides observability and AIOps capabilities for Prometheus, including PromQL instant and range queries, target and rule health checks, alert inspection, TSDB status, and governed write operations such as creating silences and reloading the Prometheus configuration.
Observability AIops
Audit availability: Real writes require a committed
pendingaudit row and stop before contacting the platform if storage is unavailable. Reads and previews may continue with a warning; new audit rows store outcome summaries rather than full responses.
Disclaimer: Community-maintained open-source project. Not affiliated with, endorsed by, or sponsored by the Prometheus or Grafana projects, Grafana Labs, or the Cloud Native Computing Foundation. Prometheus, Alertmanager and Grafana are trademarks of their respective owners. MIT licensed.
Governed AI-ops for a self-hosted observability stack in one server —
Prometheus (HTTP API, PromQL, targets, rules, alerts), Alertmanager
(alerts + silences), Grafana (dashboards, datasources, folders), and
Grafana Loki (bounded LogQL log reads + log RCA) — with a built-in
governance harness: unified audit log, token/runaway budget
guard, undo-token recording, and descriptive risk-tier labels. One config can
span your whole stack; each target names its own platform.
Beyond the mock test suite, the Prometheus/Alertmanager/Grafana reads, the RCAs,
and the governed silence + dashboard write paths (with undo) have been exercised against a live Prometheus 3.x + Alertmanager + Grafana 13 stack — see
docs/VERIFICATION.md.
This is the self-hosted-observability complement to enterprise monitoring suites: it speaks the open Prometheus/Grafana APIs an SRE actually runs, not a vendor NMS.
What it does
Answers the questions an SRE actually repeats over a Prometheus/Grafana stack, and guards the writes that follow:
PromQL + metadata — instant and range queries, label-value enumeration, and series metadata, all read-only and result-capped.
Scrape-target & rule health — which targets are up/down (and why, from
lastError), which were dropped by relabeling, and which recording/alerting rules are erroring.Alerts & silences — firing/pending Prometheus rule alerts, Alertmanager's post-routing view, and its silences.
Grafana — dashboards, datasources (+ health), and folders.
Loki logs — bounded LogQL reads (label + label-value enumeration, a validation-gated
query_range, and a canned error-tail), all read-only with a hard lookback + line cap, optional multi-tenantX-Scope-OrgID, and basic/bearer auth per target.Flagship analyses — transparent heuristics that show their numbers:
firing_alert_rca(join each firing alert to its rule expr → cause + action),target_scrape_health_analysis(rank down/erroring scrapes → likely cause),alert_noise_and_flap_analysis(frequently-repeated / duplicate alerts → dedup/rollup recommendation), plus two log analyses —log_error_burst_rca(per-stream error burst vs baseline → new-signature / volume-spike / single-instance / no-baseline) andlog_volume_analysis(top streams + high-cardinality label warnings + retention hint) — andalert_log_context, which correlates a firing Prometheus alert to its Loki streams.Governed writes — create/expire Alertmanager silences (time-boxed), create Grafana annotations, update/delete dashboards, and hot-reload the Prometheus config — each audited, risk-tiered,
dry_run-able, and the reversible ones capture the real fetched before-state for undo.
Related MCP server: cicd-aiops
What this tool does, and does not, decide
It delivers Prometheus + Grafana operations — reads and writes — accurately and efficiently, and records every one of them. It does not decide whether a write is allowed to happen. That is the agent's judgement, or the permission of the account you connect it with: give it a Grafana token with only Viewer scope, and a Prometheus/Alertmanager reached without the admin/write API, and the writes fail at the server — the place that actually owns the permission.
So there is no read-only switch, no policy file, no approval gate to configure. The one thing the
tool guarantees is that nothing is silent: every call, over MCP and over the CLI alike, lands
an audit row in ~/.observability-aiops/audit.db, and destructive writes still capture their
before-state and record an inverse where one exists.
Each tool declares a
risk_level, carried into the audit row as a descriptive tier (none/confirm/review) — so a reviewer can see at a glance that a row was a high-risk delete. It is a label, not a gate.
Capability matrix (39 MCP tools)
Group | Platform | Tools | Count | R/W |
Metrics | Prometheus |
| 4 | read |
Targets | Prometheus |
| 3 | read |
Status | Prometheus |
| 2 | read |
Rules | Prometheus |
| 2 | read |
Alerts | Prometheus/Alertmanager |
| 4 | read |
Grafana | Grafana |
| 5 | read |
Loki | Loki |
| 4 | read |
Overview | all |
| 1 | read |
Analyses | Prometheus |
| 3 | read |
Log analyses | Loki |
| 2 | read |
Cross-signal | Prometheus + Loki |
| 1 | read |
Writes | Alertmanager |
| 2 | write (med) |
Grafana |
| 1 | write (medium) | |
Grafana |
| 1 | write (med) | |
Grafana |
| 1 | write (high) | |
Prometheus |
| 1 | write (med) | |
Undo | all |
| 1 | read |
all |
| 1 | write (med) |
Loki is read-only — Loki exposes no safe operational write surface (no silence/annotation analogue), so this tool deliberately ships no Loki writes.
The CLI exposes a convenience subset (query, logs, alert, overview, …);
the full 39-tool surface is via the MCP server.
Quick start
As a Claude Code plugin
One install gives an agent both the skill and the MCP server:
/plugin marketplace add AIops-tools/marketplace
/plugin install observability-aiops@aiops-toolsThe MCP server is fetched with uv and pinned to the
package version this plugin declares, so an audit row can be traced back to the
code that wrote it. Credentials are still configured with observability-aiops init — see below.
As an OpenClaw plugin
The same bundle is published on ClawHub, where one install delivers the skill and its MCP server together:
openclaw plugins install clawhub:@zw008/observability-aiops
openclaw skills info observability-aiops # expect: Visible to model: yesRestart the OpenClaw gateway afterwards so it loads the plugin. The MCP server is
fetched with uv, pinned to this exact release, so
uvx has to be on PATH — without it the skill still installs but reports
Visible to model: no. Credentials are configured exactly as below.
As a CLI or standalone MCP server
uv tool install observability-aiops # or: pipx install observability-aiops
observability-aiops init # wizard: pick platform (prometheus/grafana) + store the token (encrypted)
observability-aiops doctor # verify config, secrets, connectivity
observability-aiops overview # snapshot: firing alerts + targets up/down + rules erroring
observability-aiops query instant 'up' # run a PromQL instant query
observability-aiops logs errors '{app="api"}' # tail error-level Loki logs (bounded)
observability-aiops alert rca # root-cause the firing alertsRun as an MCP server (stdio):
export OBSERVABILITY_AIOPS_MASTER_PASSWORD=... # unlock secrets non-interactively
observability-aiops mcpWhere that password then lives: an exported variable is readable by every process this shell starts and is recorded by shell history. On a shared or long-lived host, prefer the interactive prompt, or inject it from a secret manager for the life of the one command that needs it.
Governance
Every MCP tool passes through the bundled @governed_tool harness:
Audit — every call (params, outcome summary, status, duration, risk tier, and any operator-supplied approver/rationale) is logged to
~/.observability-aiops/audit.db(relocatable viaOBSERVABILITY_AIOPS_HOME). The CLI writes the same row the MCP path does — there is no unaudited entry point.Runaway guard — a safety backstop, not an authorization gate: the same call hammered in a tight loop trips a circuit breaker. Disable with
OBSERVABILITY_RUNAWAY_MAX=0; optional hard ceilings viaOBSERVABILITY_MAX_TOOL_CALLS/OBSERVABILITY_MAX_TOOL_SECONDS.Undo recording — reversible writes record an inverse descriptor built from the fetched before-state (
create_silence→expire,update_dashboard/delete_dashboard→restore the captured prior model).Risk tier — a descriptive label on the audit row derived from
risk_level; it gates nothing.
Supported scope & limitations
Platforms: Prometheus HTTP API (+ a companion Alertmanager), Grafana HTTP API, and Grafana Loki HTTP API (read-only). Hosted/SaaS monitoring suites (Datadog, New Relic, enterprise NMS) are deliberately out of scope for this tool.
Verification. The mock suite covers all four platforms; in addition the Prometheus, Alertmanager and Grafana surfaces have been exercised against a live Prometheus 3.x + Alertmanager + Grafana 13 stack (RCAs, the silence and dashboard governed writes, and undo replay). The Loki surface has not yet been exercised live. All four are free and open-source and trivial to stand up in a lab (
docker run prom/prometheus,grafana/grafana,grafana/loki), soobservability-aiops doctoris the fastest live check (Prometheus/api/v1/status/buildinfo, Grafana/api/health, Loki/ready+/loki/api/v1/status/buildinfo). Seedocs/VERIFICATION.md.
Missing a capability?
Want another read, an analysis tuned, or a platform capability that isn't here? Open an issue or a PR — feedback and contributions are welcome.
Available Tools
39 toolsalert_log_contextARead-only
[READ][cross-signal] Correlate a firing Prometheus alert to its Loki log streams.
Reads the firing alert's labels from the Prometheus target, maps the Loki-friendly ones (namespace, job, service, app, container, pod, instance, component — first four, in that priority order) into a LogQL stream selector, and pulls the correlated error streams from the Loki target. Best-effort: label values are escaped into the selector and only labels the alert and Loki actually share will match.
Args: alertname: The firing alert's name (from firing_alerts / firing_alert_rca). hours: Lookback window in hours (capped at the tool's max lookback). target: Prometheus target name from config; omit for the default. loki_target: Loki target name from config; omit for the default.
Inputs: alertname (required), hours, target, loki_target.
alertmanager_alertsBRead-only
[READ] Alerts as Alertmanager sees them (post grouping/silence/inhibit).
Args: active_only: If True, exclude silenced/inhibited alerts. target: Prometheus target name from config (its Alertmanager); omit for default.
Inputs: target, active_only.
alert_noise_and_flap_analysisBRead-only
[READ][analysis] Find noisy/duplicate alerts → dedup/rollup recommendation.
Args: noise_threshold: Instance count at/above which an alertname is "noisy". target: Prometheus target name from config; omit for the default.
Inputs: target, noise_threshold.
create_annotationADestructive
[WRITE][risk=medium] Create a Grafana annotation (an event marker).
Args: text: Annotation text. tags: Optional list of tag strings. dashboard_uid: Optional dashboard UID to attach the annotation to. dry_run: If True, preview without creating. target: Grafana target name from config; omit for the default.
Inputs: text (required), tags, target, dry_run, +1 more.
create_silenceADestructive
[WRITE][risk=medium] Create a time-boxed Alertmanager silence. Inverse: expire_silence.
Args: matchers: List of {name, value, isRegex?} to match alerts to silence. minutes: Silence duration in minutes (must be > 0; silences are time-boxed). comment: Reason recorded on the silence. dry_run: If True, preview without creating. target: Prometheus target (its Alertmanager); omit for the default.
Inputs: matchers (required), target, comment, dry_run, +1 more.
datasource_healthBRead-only
[READ] Health of one Grafana datasource.
Args: datasource_id: Numeric datasource id (from list_datasources). target: Grafana target name from config; omit for the default.
Inputs: datasource_id (required), target.
delete_dashboardADestructive
[WRITE][risk=high] Delete a Grafana dashboard. Captures the prior model BEFORE delete.
Pass dry_run=True to preview (reports the dashboard title). Optionally set an approver (OBSERVABILITY_AUDIT_APPROVED_BY) to annotate the audit row — it is not required. The prior model is captured so the recorded undo can recreate it.
Args: uid: Dashboard UID to delete (from list_dashboards). dry_run: If True, preview without deleting. target: Grafana target name from config; omit for the default.
Inputs: uid (required), target, dry_run.
dropped_targetsARead-only
[READ] Targets discovered but dropped by relabeling.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
expire_silenceADestructive
[WRITE][risk=medium] Expire (delete) an Alertmanager silence by id.
Args: silence_id: The silence id (from list_silences / create_silence). dry_run: If True, preview without expiring. target: Prometheus target (its Alertmanager); omit for the default.
Inputs: silence_id (required), target, dry_run.
firing_alert_rcaARead-only
[READ][analysis] Root-cause firing alerts: join each to its rule expr → cause+action.
Pulls firing alerts + alerting rules, matches them, and maps each to a likely cause and recommended action. Advisory heuristic — verify before acting.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
firing_alertsBRead-only
[READ] Currently firing Prometheus rule alerts, grouped by severity.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
get_dashboardARead-only
[READ] One dashboard's summary (title, version, panel + tag counts).
Args: uid: Dashboard UID (from list_dashboards). target: Grafana target name from config; omit for the default.
Inputs: uid (required), target.
instant_queryARead-only
[READ] Evaluate a PromQL expression at a single instant.
Args: query: A PromQL expression (e.g. 'up' or 'rate(http_requests_total[5m])'). time: Optional RFC-3339 or unix timestamp for the evaluation instant. target: Prometheus target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — narrow the selector or raise
the limit rather than treating it as complete.
Inputs: query (required), time, target.
label_valuesARead-only
[READ] Distinct values of a label (default name = all metric names).
Args: label: Label name to enumerate (default name). match: Optional PromQL selector to scope the values (e.g. '{job="api"}'). target: Prometheus target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — narrow the selector or raise
the limit rather than treating it as complete.
Inputs: label, match, target.
list_dashboardsARead-only
[READ] Grafana dashboards (optionally filtered by a title query).
Args: query: Optional title substring to search for. target: Grafana target name from config; omit for the default.
Inputs: query, target.
list_datasourcesBRead-only
[READ] Configured Grafana datasources (id, uid, name, type, default).
Args: target: Grafana target name from config; omit for the default.
Inputs: target.
list_foldersBRead-only
[READ] Grafana folders.
Args: target: Grafana target name from config; omit for the default.
Inputs: target.
list_rulesARead-only
[READ] All recording + alerting rules, optionally filtered by type.
Args: rule_type: Filter by "alerting" or "recording"; omit for both. target: Prometheus target name from config; omit for the default.
Inputs: target, rule_type.
list_silencesBRead-only
[READ] Alertmanager silences (active, pending, expired).
Args: target: Prometheus target name from config (its Alertmanager); omit for default.
Inputs: target.
list_targetsARead-only
[READ] Active scrape targets, optionally filtered by health (up/down).
Args: health: Filter by health state ("up" or "down"); omit for all. target: Prometheus target name from config; omit for the default.
Inputs: health, target.
log_error_burst_rcaARead-only
[READ][analysis] Root-cause an error burst: current window vs baseline, per stream.
Pulls error streams for the current window (now-hours..now) and an equal-length
baseline window (now-2*hours..now-hours), then classifies each burst as a new
error signature, a volume spike, or a single-instance burst — with a cause +
action. If the baseline window had no logs at all (retention or ingestion gap),
streams are classified no_baseline rather than new: an empty baseline is
not evidence of a regression. Advisory heuristic; verify against the raw logs.
Args: selector: A Loki stream selector to scope the analysis (e.g. '{app="api"}'). hours: Window length in hours for both current and baseline (capped). loki_target: Loki target name from config; omit for the default.
Inputs: selector (required), hours, loki_target.
log_volume_analysisBRead-only
[READ][analysis] Top streams by volume + high-cardinality (high-churn) label warnings.
Ranks the highest-volume streams under the selector, flags labels with many distinct values as cardinality-explosion risks, and adds a retention hint from Loki index stats when available.
Args: selector: A Loki stream selector to scope the analysis (e.g. '{namespace="prod"}'). hours: Lookback window in hours (capped at the tool's max lookback). loki_target: Loki target name from config; omit for the default.
Inputs: selector (required), hours, loki_target.
loki_labelsARead-only
[READ] Distinct Loki label names present in the lookback window.
Args: hours: Lookback window in hours (capped at the tool's max lookback). target: Loki target name from config; omit for the default.
Inputs: hours, target.
loki_label_valuesARead-only
[READ] Distinct values of one Loki label (bounded).
Args: name: Label name to enumerate (e.g. 'app', 'namespace'). hours: Lookback window in hours (capped at the tool's max lookback). target: Loki target name from config; omit for the default.
Inputs: name (required), hours, target.
loki_queryARead-only
[READ] Bounded LogQL query_range passthrough (validation-gated).
The query MUST carry a stream selector (e.g. '{app="api"}') — an unbounded query with no selector is rejected. Lookback is capped and the line count is clamped.
Args: logql: A LogQL query with a stream selector (e.g. '{job="api"} |= "error"'). hours: Lookback window in hours (capped at the tool's max lookback). limit: Max log lines to return (clamped to the tool's max line limit). target: Loki target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — re-run with a higher limit.
Inputs: logql (required), hours, limit, target.
loki_tail_errorsARead-only
[READ] Canned error-level read for a stream selector (error line-filter).
Args: selector: A Loki stream selector (e.g. '{app="api"}' or 'app="api"'). hours: Lookback window in hours (capped at the tool's max lookback). limit: Max log lines to return (clamped to the tool's max line limit). target: Loki target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — re-run with a higher limit.
Inputs: selector (required), hours, limit, target.
observability_overviewARead-only
[READ] Platform-aware health snapshot for the target.
Prometheus: firing-alert count + scrape up/down + rules erroring. Grafana: dashboard / datasource / folder counts.
Args: target: Target name from config; omit for the default.
Inputs: target.
pending_alertsARead-only
[READ] Pending (not-yet-firing) Prometheus rule alerts, by severity.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
prometheus_config_statusARead-only
[READ] Running-config fingerprint + size (never the raw YAML/secrets).
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
prometheus_tsdb_statusBRead-only
[READ] TSDB head cardinality stats + the top metrics by series count.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
range_queryARead-only
[READ] Evaluate a PromQL expression over a time range.
Args: query: A PromQL expression. start: Range start (RFC-3339 or unix timestamp). end: Range end (RFC-3339 or unix timestamp). step: Resolution step (e.g. '60s', '5m'). target: Prometheus target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — narrow the selector or raise
the limit rather than treating it as complete.
Inputs: end (required), query (required), start (required), step, +1 more.
reload_prometheus_configADestructive
[WRITE][risk=medium] Hot-reload the Prometheus config (POST /-/reload). No undo.
Records the pre-reload config hash. Rollback = re-apply the prior config file.
Args: dry_run: If True, preview (reports the current config hash) without reloading. target: Prometheus target name from config; omit for the default.
Inputs: target, dry_run.
rule_healthARead-only
[READ] Rule-evaluation health summary + the list of erroring rules.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
series_metadataARead-only
[READ] Series (label-set) metadata for a PromQL selector.
Args: match: A PromQL series selector (e.g. 'up{job="node"}'). start: Optional range start (RFC-3339 or unix timestamp). end: Optional range end (RFC-3339 or unix timestamp). target: Prometheus target name from config; omit for the default.
Returns an envelope with returned/limit/truncated. If
truncated is true the result is partial — narrow the selector or raise
the limit rather than treating it as complete.
Inputs: match (required), end, start, target.
target_scrape_healthBRead-only
[READ] Up/down scrape-health summary plus the list of unhealthy targets.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
target_scrape_health_analysisBRead-only
[READ][analysis] Rank down/erroring scrape targets and classify each cause.
Args: target: Prometheus target name from config; omit for the default.
Inputs: target.
undo_applyADestructive
[WRITE][risk=medium] Apply a recorded undo by dispatching its inverse tool.
The inverse runs through its own governed tool, so its real risk tier is recorded there too. Pass dry_run=True to preview the inverse call without executing it. A token can only be applied once.
Args:
undo_id: The undoId from undo_list (or an _undo_id in a write result).
dry_run: If True, preview the inverse tool + params without running it.
target: Passed through to the inverse tool when it accepts a target.
Inputs: undo_id (required), target, dry_run.
undo_listARead-only
[READ] List recorded, not-yet-applied undo tokens (most recent first).
Each entry names the original tool, the inverse tool that undo_apply
would run, and a human note. Use the undoId with undo_apply.
truncated is measured (one extra row is fetched), not guessed from a
length coincidence — when it is true, re-run with a higher limit.
Each entry carries effectVerified. False means the original write
lost its response, so the change it reverses is PROBABLE, not confirmed —
check the live state before applying, and do not report the result as a
restore of a state that may never have been reached.
Args: limit: Max rows to return (default 50). target: Unused (undo state is host-local); accepted for CLI uniformity.
Inputs: limit, target.
update_dashboardADestructive
[WRITE][risk=medium] Update a Grafana dashboard. Captures the prior model for undo.
The dashboard model must include its uid. The current model is fetched
first and stashed so the harness records a restore undo.
Args: dashboard: Full dashboard model (must include "uid"). overwrite: Overwrite the existing dashboard version (default True). dry_run: If True, preview without updating. target: Grafana target name from config; omit for the default.
Inputs: dashboard (required), target, dry_run, overwrite.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 39 tools
Most tools have clearly distinct purposes, well separated by backend (Prometheus/Loki/Grafana/Alertmanager) and by raw-read vs analysis. Minor overlaps exist in the scrape-health cluster (list_targets vs target_scrape_health vs target_scrape_health_analysis) and between loki_query and the canned loki_tail_errors, but the descriptions distinguish raw vs summarized vs analyzed views reasonably.
Names are consistently snake_case, with verb prefixes for mutations (create_/delete_/update_/expire_/reload_/undo_) and analysis-suffixed diagnostics. Many read/state tools are noun-phrases (firing_alerts, prometheus_tsdb_status, instant_query) rather than verb_noun, which is a minor deviation but internally predictable.
39 tools is heavy and well above the comfortable range, raising selection cost. However, the server legitimately spans four backends plus a cross-signal analysis layer and an undo/governance layer, so most tools do map to a real capability rather than being redundant.
Coverage is strong: full silence lifecycle (create/list/expire), alerts, rules, scrape targets, Loki queries, PromQL queries, dashboard read/update/delete, and a rollback layer. The notable gap is the absence of a create_dashboard tool (only update/delete exist), plus datasource/annotation operations are mostly read-only.
Maintenance
Related MCP Connectors
Autonomous and governed SRE & MLOps control plane for AI agent platforms.
Governed MCP gateway: one endpoint for your tools, with credential custody and audit log.
MCP observability. Query live traffic, errors, duration, and alerts from your AI agent.
MCP-native AI SRE: ask what's broken in production, get a reviewed GitHub fix PR.
Related MCP Servers
- AlicenseAqualityBmaintenanceGoverned PostgreSQL DBA operations — slow-query, bloat, and blocking-lock RCA, index management, vacuum/analyze, and replication lag, with unbypassable audit logging (MCP + CLI), budget/runaway guards, dry-run, and undo/rollback.35MIT
- AlicenseAqualityBmaintenanceGoverned CI/CD operations for self-hosted GitLab and Gitea — pipeline-failure, runner, artifact-bloat, and stale-branch RCA, with unbypassable audit logging (MCP + CLI), budget/runaway guards, dry-run, and undo/rollback.28MIT
- AlicenseAqualityBmaintenanceGoverned MinIO object-storage operations — capacity, exposure-surface audit, lifecycle, and healing RCA, with guarded bucket writes, unbypassable audit logging (MCP + CLI), budget/runaway guards, dry-run, and undo/rollback.48MIT
- AlicenseAqualityBmaintenanceGoverned reverse-proxy operations for Traefik, Caddy, and HAProxy — backend-health, cert-expiry, 5xx, and route-conflict RCA, with unbypassable audit logging (MCP + CLI), budget/runaway guards, dry-run, and undo/rollback.28MIT