Using AI coding agents

Set up, customise and extend Open Core HR with Claude Code, Codex, Cursor, GitHub Copilot or any coding agent.

Use this page with AI Open in Claude Open in ChatGPT View as Markdown

Open Core HR is built to work with AI agents. Installation with an agent is free (see Install with an AI agent). This page covers what the download gives your agent, how to connect it to always-current documentation, and the safety rules it follows.

What is in your download

Open the folder where you extracted the Core download (the guides call it backend/) in your agent. It reads these automatically:

File	What it gives the agent
`AGENTS.md` (and `CLAUDE.md`)	The architecture, module system, commands, and the rules that keep updates safe
`INSTALL-WITH-AI.md`	The step-by-step install and first-run checklist
`.claude/skills/`	Ready-made skills: setup, create-module, customize, update-and-backup, troubleshoot
`.mcp.json`	A connection to the Open Core HR documentation server

Agents other than Claude Code read AGENTS.md; point them at the skill files in .claude/skills/ when you ask for those tasks.

Useful first prompts:

"Install Open Core HR on this server using INSTALL-WITH-AI.md."
"Run php artisan doctor and fix anything it reports."
"Create a module called AcmeTimesheets for weekly timesheets."
"Change the company branding and add French translations."

Connect the documentation server (MCP)

The Open Core HR MCP server gives your agent the current documentation, add-on catalogue, installation plans and release notes, so it does not rely on what it remembers. It is read-only and needs no account.

Address: https://czappstudio.com/mcp

Claude Code (already configured by the .mcp.json in your download):

claude mcp add --transport http open-core-hr https://czappstudio.com/mcp

Cursor, VS Code and other MCP clients: add a remote (HTTP) MCP server with the address above, for example in .cursor/mcp.json:

{ "mcpServers": { "open-core-hr": { "url": "https://czappstudio.com/mcp" } } }

Codex (~/.codex/config.toml):

[mcp_servers.open-core-hr]
url = "https://czappstudio.com/mcp"

Without MCP, agents can read llms-full.txt, which contains every guide as plain text.

Once connected you can also simply ask questions, for example "How do leave balances carry over?" or "Which add-on handles customer visits?".

Safety rules your agent follows

The agent files tell agents to:

Check the environment first. When APP_ENV=production, the agent will not run commands that delete or reset data (migrate:fresh, migrate:refresh, migrate:reset, db:wipe, demo seeders, deleting storage or Modules, changing APP_KEY) unless you clearly confirm you want that on production. Open Core HR itself also refuses the database reset commands in production.
Back up before risky changes (see Backups and updates) and prefer trying changes on a copy.
Keep changes update-safe: build features as your own module, never edit Core or store add-on files (see Building your own module).
Never put secrets in code, commits, prompts or support tickets.

In Claude Code, a hook in .claude/settings.json enforces the production rule even if the agent forgets.

Ready-made add-ons and services

When you ask for a feature that an existing add-on already provides, the agent mentions it once with a link, so you can choose between buying it and building your own; it builds your own if you prefer. If installation keeps failing, it may mention CZ App Studio's installation plans, where an engineer does it for you. Prices and plans always come from the website.