Connecting your AI
Connect Claude, ChatGPT or any MCP app to your servers, what it can and cannot do, and how you stay in charge.
You can run your servers by talking to your own AI app: install a plugin, change a setting, apply a premade setup, or explain a crash. It works through a connector we host, on every plan. Your AI app pays for the AI; connecting costs nothing.
Connect it
You add one connector for your whole account, not one per server. Its address is
https://mcp.quarry.host/mcp, also shown on the Connect your AI page.
- Claude: web, desktop and phone.
- ChatGPT: plans that allow custom connectors.
- Claude Code, Cursor and other MCP apps.
In every app: add a custom connector with the address, sign in to Quarry Hosting, then check the consent screen. In your AI app the connector is named Quarry Hosting.
The consent screen
- Which app is asking. Apps we do not recognise are marked Unverified app.
- Sends you back to: the full address your answer goes to. For Claude it is
https://claude.ai. If you do not know the address, choose Don't connect. - Your servers and what it may do on each. Open a server to change that before you connect.
Name the connection if you like, then choose Connect. We email you about every new connection. Disconnect any time in Account, Connected AIs.
You stay in charge
- Each action on each server is set to Allow, Ask or Block. New servers start on the Balanced preset: risky actions such as ops, ranks, plugins, resets and rollbacks ask first.
- A snapshot is taken before every change, and every step has Undo.
- Stop pauses your AI on a server at once.
See The AI page for the full table, approvals and undo.
What your AI can do
Every call names a server, and that server's settings apply.
- Plugins and settings: search and install plugins from Modrinth and Hangar, change settings, config files and gamerules. Every jar passes our safety check.
- Ranks: make LuckPerms ranks and give permissions. The
*permission stays with you. - Explain and fix: read logs and crash reports, run a diagnosis, suggest or make a fix.
- Schedules, premade setups, worlds and versions: make schedules, apply or tune a setup ("make it a lifesteal server with 5 hearts"), reset a world, switch version.
- Commands: a limited list (messages, time, weather, gamemode, teleport, give and the like). It
cannot op, ban, run
/executeor change permissions by command. You can run anything in your own Console. - Servers: make a server in your free slot.
- Docs: search these pages.
There is no owner API for your own scripts, so there are no API keys or call limits to look up. Use the panel or your AI.
What your AI can never do
- Approve its own requests, change its permissions or turn on chat reading.
- Spend credits: buy, change a plan or boost.
- Touch your team, account, alerts, SFTP logins, custom domain, internet allowlist or region.
- Delete a server or download backups.
- See passwords, keys or tokens. Secrets in config files show as
[redacted: secret].
Anything players wrote (chat, signs, books, names, logs) reaches your AI marked as untrusted. Your AI cannot read chat at all unless you turn that on.
Team members
Co-owners and admins can connect their own AI to a server they help run. It can only do what their role allows, and the server's settings still apply. Removing someone cuts their AI off at once.
If something goes wrong
If your AI says it cannot do something, check that server's settings on the AI page, then your approvals. If it says it was stopped, someone pressed Stop. See Troubleshooting.