measure-monitor 0.1.0

Unattended cross-platform anomaly monitoring for the Measure ecosystem: fetch metric observations, detect irregular drops/spikes, deliver alerts.

Measure Monitor

Unattended cross-platform anomaly monitoring for the Measure ecosystem.

measure-monitor is the ecosystem's orchestrator, not an MCP product. It sits one level above the products — which never import one another — and wires them together: it fetches metric Observations from a source product (paid media today), runs the SDK's shared anomaly detector, and delivers alerts to your channels. Run it on a schedule for hands-off monitoring of irregular drops and spikes.

  ┌── cron / /schedule ──────────────────────────────────────────────┐
  │  measure-monitor run --config monitors.toml                       │
  │                                                                   │
  │   paid-media.get_metric_observations(metric)   ← fetch + history  │
  │            │                                                      │
  │            ▼  detect_anomaly (SDK)              ← shared brain     │
  │       irregular drop/spike?                                       │
  │            │                                                      │
  │            ▼                                                      │
  │     Slack / Teams / email                       ← centralized     │
  └───────────────────────────────────────────────────────────────────┘

Install

pip install measure-monitor          # pulls in measure-paid-media-mcp + measure-sdk
measure login                        # one-time Google sign-in (from the SDK)

Configure

Describe what to watch in a JSON or TOML file — see examples/monitors.toml:

[[monitors]]
name = "spend-anomaly"
metric = "spend"        # spend | revenue | conversions | clicks | impressions | roas | cpa | ctr
days = 7                # each period is a 7-day window
periods = 4             # 4 prior windows form the baseline
method = "z_score"      # z_score | mad | pct_change
sensitivity = 3.0       # stdevs (z_score/mad) or fraction (pct_change)

  [[monitors.channels]]
  type = "console"      # console | slack | teams | discord | email

Run

measure-monitor run --config monitors.toml            # detect + deliver
measure-monitor run --config monitors.toml --dry-run  # detect only, no delivery

Output is a JSON report. Exit code is 1 when any anomaly was found (0 = clean, 2 = error), so a cron wrapper can react to it.

Email delivery reads SMTP from the environment: MEASURE_SMTP_HOST, MEASURE_SMTP_PORT, MEASURE_SMTP_USER, MEASURE_SMTP_PASSWORD, MEASURE_SMTP_FROM.

Schedule it

System cron — every hour:

0 * * * * cd /path/to/work && measure-monitor run --config monitors.toml >> monitor.log 2>&1

Claude Code /schedule — a cloud routine that runs the check and reacts to the result in natural language (no config file needed; the agent has both MCPs connected). Example prompt to schedule:

Every hour: use paid-media get_metric_observations for spend and roas across all platforms; for any anomaly, deliver a summary with the alerts engine.

Both modes respect the ecosystem rule: products never import each other — only this orchestrator depends on more than one.

How it fits the ecosystem

• Source of truth for detection: measure_sdk.detect_anomaly — the same brain every product uses, so results are comparable across platforms.
• Extensible: as GA4 / SEO products expose get_metric_observations, point a monitor at them; the runner is source-agnostic (it depends only on the Observation contract).