Light Cloud AI
Light Cloud AI (LCAI) is the MCP server that lets Claude Code, GitHub Copilot, and any MCP client sign you up, deploy, manage and pay for your apps - and the Light Cloud CLI does the same from a shell. Install once, approve a code from any device, and deploy by asking.
claude mcp add light-cloud -- npx @light-cloud/mcp-server
Then, in your project folder: “Deploy this project to Light Cloud.” Two questions, one URL. No account yet? It makes one.
One command registers it with Claude Code. Node.js 18 or later is all you need; npx fetches the package on first run.
claude mcp add light-cloud -- npx @light-cloud/mcp-server
Open Claude Code in your project folder and ask in plain language. Claude asks at most two things, in one message: which workspace, if you have several, and whether the site should be public or behind a password. No account yet? Say “connect me as you@example.com”: Claude shows a short code, you approve it on any device, and the account exists. No password, no form. Credentials stay on your machine in ~/.lightcloud.
Deploy this project to Light Cloud.
Claude uploads the folder; Light Cloud reads it with the same detector the console uses for a repository, builds, and Claude waits for the build. Claude Code asks before each tool call; to approve them once, add mcp__light-cloud under /permissions.
The reply is the live URL on light-cloud.io - and the password, if you asked for one - plus a console link. A small .lightcloud file records the app, so follow-ups need no arguments.
Redeploy. Then show me the logs.
$ claude mcp add light-cloud -- npx @light-cloud/mcp-server
$ claude
Install, ask, approve, live — the whole loop in one terminal.
The same cloud, as commands.
The Light Cloud CLI reaches every feature the MCP server does - sign-up, deploys, logs, databases, plans, the card - as commands with --json output. For scripts, CI, and agents that already know how to run a terminal.
npx @light-cloud/cli login --device
Sign in - or sign up - with a code you approve on any device. No browser needed here.
lc init && lc deploy
Link the folder to an app (existing or new) and follow the build to a live URL.
lc billing card add --plan pro
A Stripe-hosted link saves the card; the plan switches when it lands. Card numbers never touch the terminal.
lc logs -f --json
Every command takes --json, so an agent or a pipeline can drive it like git.
Things you can just ask.
No tool names to remember. The assistant picks the right calls and chains them, so “why did it fail?” looks up the deploy, reads the logs, and summarizes.
“Deploy this project to Light Cloud.”
“Create an app from my acme/storefront repo and deploy the main branch.”
“Create a staging environment for storefront and deploy the develop branch to it.”
“Why did the last deployment of storefront fail? Check the logs.”
“Connect me to Light Cloud as julia@example.com.”
“Show me the plans, add a card, and put the workspace on Pro.”
“Add a Postgres database and set DATABASE_URL on production.”
“Deploy this project behind a password.”
“Redeploy.”
Handled for you.
The parts you would otherwise do in a dashboard - the account, the plan, the card, the build settings - happen in the conversation or are read from the folder you are in.
An email and a short code, approved on any device, create the account and a free workspace. No password ever enters the chat.
Pick a plan from the conversation. The card goes through a Stripe-hosted link the tools wait on - never through the tools themselves.
A missing size, an exhausted free pool or a missing card come back naming the tool that fixes it, so the agent keeps going.
Create a database, set its connection string and the other variables, scale to always-on - each a sentence, not a form.
The uploaded folder is read by the same detector the console uses for a repository - Next.js, React, Vue, Python, Go, Rails, Dockerfiles and forty more - so an upload and a repo deploy the same way.
Variables in your local .env are set on the app for you.
Every deploy comes back with a light-cloud.io URL and a console link. Ask for a password and visitors are asked for it at the edge, from the first request.
A .lightcloud file records the app, so later requests run without repeating yourself.
Good to know.
Install LCAI or the CLI, ask for a deploy, and you are live - account, plan and card included. The docs cover Claude Desktop, a custom API endpoint, and troubleshooting.