How to Monitor Claude Status (and Get Alerts)

When Claude feels “down”, the fastest way to confirm is the official status page. The next fastest way is to have a monitor that checks for you and alerts you only when something changes.

This tutorial shows how to do both on Zo Computer:

  • Check the official Claude status page quickly

  • Set up a Zo Agent that polls the status API on a schedule

  • Store the last known state in a file (so the Agent has “memory”)

  • Get notified when Claude goes from Operational → Degraded → Partial outage, etc.

Prerequisites

  • A Zo Computer workspace

  • A place to receive alerts (email or SMS)

Useful references:

  • Claude status page (official): https://status.anthropic.com/^1

  • Alternate Claude status page: https://status.claude.com/^2

  • Zo Agents docs: https://docs.zocomputer.com/agents^3

  • Zo web reading tool (fast extraction): https://docs.zocomputer.com/tools/read-webpage^4

Step 1: Decide what you want to monitor

Claude “being down” can mean different things:

  • The consumer app (claude.ai) is failing

  • The developer console (platform.claude.com) is failing

  • The API (api.anthropic.com) is returning errors or overloads

The status pages above usually track these as separate components. The monitoring approach below works for any of them.

Step 2: Pick a stable machine-readable endpoint

Most public status pages expose a simple JSON endpoint.

For Anthropic’s status page, the common endpoint is:

  • https://status.anthropic.com/api/v2/summary.json^5

That endpoint is ideal for monitoring because:

  • It’s structured (easy to diff)

  • It changes only when the status changes

  • It doesn’t include lots of layout/HTML noise

Step 3: Create two small “memory” files in your workspace

Create a folder like:

  • file 'Monitoring/claude-status/'

And inside it, keep:

  • last-summary.json — the last successful status payload

  • last-seen.txt — a short, human-readable last known status (what you alert on)

The key idea: the Agent reads these files, checks the current status, compares, and only alerts on meaningful change.

Step 4: Create a Zo Agent that checks Claude status on a schedule

Create an Agent that runs every 5 minutes (or every 1 minute if you care a lot about outage response).

Use an instruction like this (copy/paste as your Agent instruction):

  1. Fetch https://status.anthropic.com/api/v2/summary.json.

  2. Parse out:

    • overall page status (e.g. "operational")

    • any active incidents (name + status)

    • any components that are not operational

  3. Read file 'Monitoring/claude-status/last-seen.txt' if it exists.

  4. If the “human summary” changed:

    • Email me the new summary

    • Include the most relevant incident/component details

  5. Write the new summary into file 'Monitoring/claude-status/last-seen.txt'.

  6. Also write the full raw JSON into file 'Monitoring/claude-status/last-summary.json'.

Notes:

  • This is intentionally “diff-driven”. You don’t want a notification every 5 minutes; you want a notification when status changes.

  • If you prefer SMS for outages, use SMS delivery for the Agent, but keep the message short (one line summary + link to the status page).

Step 5: Make it resilient (avoid false alarms)

Two common failure modes:

  1. The status endpoint is temporarily unreachable from your network.

Fix: treat network fetch failures as “unknown”, but don’t alert unless you see repeated failures.

  1. Claude is “up” but your specific account is rate-limited or overloaded.

Fix: add a second check that does one lightweight API call (if you use the API) and only alert when that call fails consistently.

Step 6: When you need a screenshot (optional)

Sometimes you want to capture what the status page looked like at the time of the incident.

In those cases, use the browser-rendered reading tool:

  • https://docs.zocomputer.com/tools/view-webpage^6

It’s slower, but it can capture what you’d see in a normal browser.

Summary

You now have a practical “status monitor” pattern on Zo:

  • Source of truth: Claude’s official status page

  • Durable memory: a small state file in your workspace

  • Execution: a scheduled Agent

  • Output: an alert only when something changes

This same pattern works for any service that has a public status page (OpenAI, GitHub, AWS, etc.).