Most alerting setups fail for the same reason: someone picks a round number like 80% CPU, hopes for the best, and ends up drowning in false positives or missing real incidents. What if your AI assistant could look at your actual metrics history and set thresholds that make sense?

That's exactly what you can do now with Quave ONE MCP and our new Alerts feature - together.

Here's what the conversation looks like inside your IDE or preferred place to run MCPs:

"What's been the average CPU usage for production over the last 7 days?"

Your AI assistant pulls the metrics series, sees that production typically sits at 35-45% CPU with occasional spikes to 62% during peak hours. No guesswork needed.

"Set up a CPU alert at 70% for 5 minutes -- and a P95 latency alert at 3 seconds. Send both to our Slack ops channel."

Done. Two alerts configured with thresholds grounded in real data, not gut feelings. The contact point is wired up. If something actually goes wrong, your team knows.

That's the power of combining an AI assistant that can read your infrastructure history with an alerting system flexible enough to act on what it finds.

Another example is to investigate issues like build failures or restarts of containers. Watch it in action here.

---

Quave ONE MCP

The MCP (Model Context Protocol) gives your AI assistant direct, controlled access to your Quave ONE infrastructure from inside your IDE. No more context-switching between dashboards, terminals, and docs.

With 59 tools (and probably growing), it covers the full lifecycle: creating apps, deploying, scaling, viewing logs and metrics, managing environment variables, configuring alerts, rolling back, and even searching our documentation - all through natural language.

It works with Cursor, Claude Desktop, Claude Code, and any MCP-compatible client. No self-hosting required - just connect to https://mcp.quave.cloud/ and you're ready.

Security Built In

We designed the MCP with security as a first-class concern:

• Server-enforced scopes -- Control exactly what each MCP key can do: read-only, configuration, deployment, or dangerous operations. Even a compromised client cannot exceed the permissions you granted.
• Secrets isolation -- Viewing secret environment variables requires a separate scope.
• Encrypted keys with per-key scope control and last-used tracking for auditing.

You stay in control of what your AI assistant can and cannot do.

Please read our MCP Docs to see all we offer.

---

Alerts Are Now Generally Available

Set up monitoring rules for your environments and get notified the moment something needs attention.

What You Can Monitor

• Resource usage -- CPU and memory, both as combined totals across all containers or per individual pod. Per-pod monitoring is great for spotting an overloaded database primary while replicas sit idle.
• HTTP traffic -- Request rate, 5xx and 4xx error rates and percentages, P95 and P99 response times.
• Traffic drops -- Set a "less than" threshold on request rate to detect when a bad deploy kills your traffic.

BTW, pods and containers are the same thing, we should probably consolidate on pods sometime in the future :), sorry for the confusion.

Flexible Configuration

Each alert is fully configurable: choose a metric, set a threshold and operator, pick how long the condition must hold before firing (from 1 minute to 1 hour), and select an aggregation method (average, max, min, or last value). You can also configure the query time window for how far back the system looks at data.

Notify Your Team Where They Already Work

Alerts are delivered through Contact Points -- notification destinations you set up at the account level:

• Slack -- with channel and user mention support
• PagerDuty -- with configurable severity
• Webhooks -- for custom integrations
• Email -- to one or more addresses

A single alert can notify multiple contact points simultaneously - Slack for visibility and PagerDuty for on-call escalation at the same time.

Please read our Alerts docs to understand nuances.

---

Better Together

The real value, IMO, is the combination. Use MCP to pull your metrics history, understand your baseline, and configure alerts with thresholds that reflect how your app actually behaves - not arbitrary defaults. And when an alert fires, ask your AI assistant to check the logs, inspect pod health, or roll back, all without leaving your editor.

The power of using these tools inside your editor/workspace is that it knows your code and can propose solutions and even fix the issues immediately (and then deploy them and check results. I do this everyday since December of 25, it's amazing).

Both features are available today. Head to your Quave ONE dashboard to set up alerts inside your app envs, or check our MCP documentation to connect your AI assistant.

---

Other Improvements and Fixes

While MCP and Alerts were the headline features, we shipped a bunch of other improvements over the past couple of weeks:

• Meteor.js 3.4 preset - Our Docker preset now supports Meteor 3.4 out of the box.
• Persistent storage for all users - Disk (volume) support is no longer restricted. Any user can now deploy apps with persistent storage.
• Kafka and OpenSearch dashboards - Added built-in observability dashboards for Kafka and OpenSearch workloads.
• Smarter health checks - HTTP health probes are now conditionally applied based on your enableHttpProbes flag, so environments that don't need them won't get false-negative restarts.
• Improved Slack notifications - Alert notifications now include the account name and a direct link to the app environment, so you can jump straight from Slack to the right place.
• Mobile-friendly Alerts page -- The alerts UI is now responsive and usable on smaller screens.

Have a great weekend! Enjoy with your loved ones.

Sorry for the long text, too much to share.