Tell Claude what you need. It runs the Norbix CLI and ships a backend.
A coding agent runs commands all day. Instead of loading hundreds of tools into its context, it runs norbix like you do and reads one JSON answer per command. The MCP server stays there as the second way in.
norbix login → norbix ai init → norbix … --json --dry-run first, --yes only for destructive changes. Same commands in CI.https://hub.norbix.ai/v3/account/mcp The blocks below are the ones Cloud shows for the managed service. On your own installation Cloud fills in your own addresses.
In your project folder. The CLI is what a coding agent does best: run a command, read one JSON answer.
npm i -g @norbix.ai/cli
norbix login
norbix ai initnorbix login opens Cloud. Check the code matches your terminal, pick the roles Claude gets (for example Viewer on one project, or Administrator), press Allow.
norbix ai init wrote .claude/skills/norbix/SKILL.md and a short block in CLAUDE.md. Claude now knows to use norbix with --json, --dry-run first, and --yes only for destructive changes. You can even skip norbix login: when Claude finds it is not signed in, it shows you the link and the code and waits for your Allow.
The hosted MCP server, no key in any file. Claude signs in through the browser.
claude mcp add --transport http norbix https://hub.norbix.ai/v3/account/mcpStart claude, run /mcp, pick norbix and allow in the browser. claude mcp list shows “Needs authentication” until you do.
Put an nbsu_ key in NORBIX_API_KEY. Keep this server in the local scope: never --scope project, which writes the key into the shared .mcp.json.
claude mcp add --transport http norbix https://hub.norbix.ai/v3/account/mcp --header "Authorization: Bearer $NORBIX_API_KEY"export NORBIX_API_KEY=nbsu_…
export NORBIX_PROJECT_ID=pr_…
norbix db count orders --jsonClaude Code runs norbix whoami --json, then the read commands it needs, and answers from the JSON. Ask for a change and it shows you the --dry-run request first. With Viewer, the change is refused by the server; with Administrator, it goes through.
Which tables does my Norbix project have?
Create a project, enable modules, build collections from a schema bundle and publish them.
Find, count, insert and update records; taxonomies, aggregates and triggers on every collection.
Email, push, SMS and payment providers; templates, campaigns, files and storage.
Roles and policies, sign-in integrations, invitations and the membership settings.
Triggers across modules, scheduler tasks, serverless code functions and webhooks.
Logs by correlation id, diagnostics, usage — the same read-only look Cloud gives you.
Sign-in happens in your browser. You tick the projects and roles the agent gets — Viewer, Developer, Data Editor, Administrator or your own.
The CLI acts as an AI service user named after your computer. Remove it in Cloud and the agent is signed out.
The agent runs --dry-run first and shows the exact request. Destructive commands refuse to run without --yes.
Every call the agent makes is in Logs with its correlation id, like any other request.
Open your avatar → Connect your AI in Cloud. It holds a prompt you can paste into any AI tool, the Hub, API and MCP addresses of your installation, and one tab per tool with the exact commands — including the key variants for CI.
Copy-paste setup for every tool, a first prompt and what it returns.
norbix login, norbix ai init, and when to choose the CLI or MCP.
The address, OAuth or a key, the consent screen, roles and environments.
Every tool group per module, which ones read, change, or ask first.
Create a free account, connect Claude Code, and ask it for your first backend change.