AI setup
Most of what goes wrong when an AI assistant deploys something is a guess: the wrong port, the wrong folder, a start command the server never runs. The panel already knows the answers for your server, so it writes them down for the assistant - and leaves every secret out.
Start with the prompt - it works with any assistant. The skill is for people who use Claude every day; the MCP server lets Claude Code, or any MCP client, read your server for itself.
On a server's GitHub or Console tab, the AI setup button copies a ready-made prompt for Claude or ChatGPT. Pick the task first, and the prompt is written for it.
On the server's Console tab, create a secret link - optionally for one task: set up, fix repo, connect database or add domain. Paste it into Claude, ChatGPT or any assistant that can open links. It shows the same server description as the copied prompt, generated fresh every time it is opened, with no passwords or tokens. Search engines are told not to index it.
https://panel.renode.space/ai/rnai_xxxxxxxx.mdDownload the renode-deploy skill - a zip for Claude Code and the Claude apps, with a guide for each runtime and a checklist the assistant works through before it calls a deploy finished.
The panel hosts an MCP server. Claude Code - or any MCP client that supports the HTTP transport - can read your server's context, fetch the setup prompts, tail the console, and read its resources and files. Restarting and writing files work only if your account has those permissions on that server. Revoke the key at any time and access stops.
claude mcp add --transport http renode https://panel.renode.space/api/client/mcp --header "Authorization: Bearer ptlc_YOUR_KEY"A prompt that says what you are trying to do gets a better answer than one that only describes the server.
Start from an empty server and an empty repository, built the way this server boots.
Make a project you already have run here: the port, the start command, the folder it expects.
The prompt carries the last lines of the console, so the assistant reads the actual error instead of guessing at one.
Wire the code to the databases already attached to the server - named, with the password left out.
Point a domain at the server and serve the project on it.
Before a deploy, the panel inspects the linked GitHub repository and shows what will break.
No AI involved - the panel reads the repository itself.
Shortened, with illustrative values. The real one is filled in from your server.
You are helping me run a project on a RE:NODE server.
Task: set up a new project
Server
- Runtime: Node.js 20
- Port: read it from the SERVER_PORT environment variable and listen on 0.0.0.0
- Code lives in /home/container
- On every start the server fetches the branch below, then runs the start command
GitHub
- Repository: your-name/your-app, branch main
- Redeploy on push: on
Databases
- MySQL database s12_app (connection details are in the panel; no password here)
Limits
- 1 GB memory, 10 GB disk
Rules
- .env is not in the repository. Tell me which variables it needs;
I will upload it through the Files tab.
- Never ask me to paste a password or a token into this chat.A prompt is something you paste into somebody else's service. So it is written as if it will be read.
Database passwords, API keys and deploy tokens are never written into a prompt or the skill. The assistant is told that they exist and where they live, not what they are.
The .env file is not in the repository, and the prompt says so. The assistant tells you which variables it needs; you upload the file yourself through the Files tab.
Open a server in the panel, go to its GitHub or Console tab and press AI setup.