5.0.0

ObserverLauncher 5.0.0

5.0.0 is a big step for the AI integration (MCP). The launcher already let an AI read and control
your server; now an AI client understands what is safe and what to do, without you explaining.

This is the plain-English release note. Developers: see the full technical changelog in CHANGELOG.md.

---

What changed

The tool count is the same (75). What changed is how an AI client sees them.

It knows which actions are safe

Every tool now tells the AI client whether it only reads, whether it can destroy something, and
whether it reaches the internet. That means a well-behaved client can run read-only checks on its
own and only stop to ask you before a change instead of asking permission for every glance at your
server status.

It has ready-made workflows

If your AI client shows a menu of prompts, you now get six of them, ready to run:

• Diagnose my server — full health check, explained in plain language.
• Tune RAM and performance — looks at your memory, JVM flags and metrics, suggests a setup.
• Explain the last crash — finds the newest crash report and tells you why, in plain words.
• Set up a Paper server — recommends version, memory and key settings for your player count.
• Audit installed mods/plugins — finds wrong-loader jars, missing dependencies and conflicts.
• Install a modpack safely — searches, shows you the plan, then installs only after you say yes.

It can pull the right context by itself

Besides the fixed resources (status, console, players…), an AI can now build a resource address for
one specific server (instance/<id>/...) or a file. so it reads exactly what it needs without
extra back-and-forth, and without switching the server you are looking at.

Why it matters

Before: you told the AI what to do, step by step.
Now: you pick a workflow, and the AI already knows the safe path, read freely, ask before writing,
and follow a proven order (diagnose -> plan -> confirm -> act)

Upgrading

Install 5.0.0 over your current version. No settings change, no migration. If you use the MCP / AI
integration, restart the app and reconnect your AI client the new capabilities are advertised at
connect time.