# Light Cloud AI > Private AI agents for your apps, and a chat that builds and deploys them, in the workspace that runs everything else. Light Cloud is a deployment platform for websites, web apps, APIs and Postgres and MySQL databases. Its AI features: the chat in the console, AI agents for your apps, and an MCP server for Claude, ChatGPT, Claude Code and GitHub Copilot. - **[The chat](https://ai.light-cloud.com/chat):** Builds and changes sites, apps, databases and agents in your workspace, and shows every step. - **[AI agents](https://ai.light-cloud.com/agents):** One address and key for your app’s AI, with the instructions and data kept on the agent. - **[MCP server](https://ai.light-cloud.com/mcp):** Claude, ChatGPT, Claude Code and GitHub Copilot run Light Cloud for you. **Which plans include it?** Every paid plan includes the chat and AI agents. On Free you see both, with a button to upgrade when you want to use them. **Where does my code go?** Into a repository in your own GitHub, GitLab or Bitbucket account, private unless you ask for a public one. Light Cloud keeps no copy. **Which AI model does the chat use?** Gemini 3.1 Pro, from Google. Your own AI agents choose from the supported models. **Can it delete things or spend money?** Only after you say yes. Deleting, changing the plan, inviting members and changing a database wait for your “Yes” button. **Are my prompts used to train models?** No. Light Cloud trains no models, and the model providers do not train on these requests without permission. **Can I use Claude or ChatGPT instead?** Yes. Connect Light Cloud’s MCP server and ask there: the same tools, on the same workspace. ## [The chat](https://ai.light-cloud.com/chat) The chat sits beside every console page. Ask for something new, a change or a fix: it does the work and shows each step. - **Builds:** Writes the code into a new repository you own, creates the app and its database, deploys and hands you the link. - **Changes:** Reads your repository before it edits, commits the change and redeploys. Your own edits stay. - **Fixes:** When a deploy fails it reads the build log, says what went wrong, fixes what it can and deploys again. - **Runs:** Status, logs, domains, databases, folders, members and plans: everything the console does. - Every step is visible: deploys with their live status, commits with their files, log excerpts. - The live app opens in a preview beside the conversation. - Questions come with buttons for the likely answers. - Deleting, paying and changing members wait for your yes. - It acts as you, with your permissions in the workspace. - **Plans:** Every paid plan, 30 messages per person per day. The chat’s own AI use is not charged; what it creates is billed like anything you create yourself. - **Your data:** Conversations and attached files are kept for 30 days. Code lives in your repository, not with Light Cloud. - **The model:** Gemini 3.1 Pro, from Google. Google does not train on these requests without permission. **Do I need to know how to code?** No. Say what you want and it chooses the framework and writes the code. Everything it writes is in your repository, to read or change. **Can it work on an app I already have?** Yes. It reads the app’s repository, makes the change as a commit and redeploys. **What can I attach?** PDF, Word, text, Markdown, CSV, JSON, HTML, XML and YAML files. They are read in your browser and kept for 30 days. **Does the Agents & CLI switch turn it off?** No. The chat is part of the console. The switch covers Claude, ChatGPT, the CLI and other outside assistants. **What happens when I reach 30 messages?** The chat says so and opens again the next day. The console works as usual in the meantime. ## [AI agents](https://ai.light-cloud.com/agents) One address and one key. The agent holds the instructions and data; your app sends the conversation and gets the answer. Apps call https://api.light-cloud.com/ai/v1 in the OpenAI Chat Completions format with the agent's key. - **Instructions:** Who the agent is and how it answers. - **Guidelines:** Rules it keeps, whatever it is asked. - **Data:** Pasted text, web pages and files, up to about 25,000 tokens, sent with every message. The page shows what that costs per message. - **Try it:** Send a test message from the console before your app does. - **In the console:** The AI agent tile on the Create page. - **From the chat:** “An agent that answers questions from my price list.” It creates the agent, adds the data and connects your app. - **From your assistant:** The MCP server’s agent tools, from Claude or Claude Code. - Tasks: name each kind of request (x-lc-task: support-reply) and each keeps its own model, usage and history. - Keep mode stays on the model and vendor you chose. - Save mode moves a task to a cheaper vendor of the same model when that saves 10% or more, and suggests cheaper models for you to approve. A model never changes without your yes. - Failover: when a vendor fails before answering, the next one takes the request. - Models: Gemini 4, Gemini 3, Gemma 4 and gpt-oss, each with its price on the models page. - **Cost:** The vendor’s price plus 5%, counted against your plan’s included usage. Your workspace’s one usage limit covers it. - **On its own:** Add a schedule and Light Cloud calls a URL on your app every 15 or 30 minutes, every hour, every 6 hours or once a day. - **Your data:** Light Cloud keeps request metadata for 30 days, never prompts or answers. The model providers do not train on these requests without permission; a provider may cache a request in memory for up to 24 hours and log prompts to detect abuse. **Do I need my own OpenAI or Google key?** No. The agent’s address and key are all your app needs. **Which SDKs work?** Any that speak the OpenAI Chat Completions format. Set the base URL and the key; streaming and tool calls work as usual. **Can my app send its own system message?** Yes. The agent’s instructions, guidelines and data go in front of it. **Which plans include AI agents?** Every paid plan. Requests from a Free workspace are refused with plan_required. **Is there a rate limit?** 600 requests a minute per key. ## [MCP server](https://ai.light-cloud.com/mcp) Light Cloud’s MCP server lets Claude, ChatGPT, Claude Code and GitHub Copilot create, deploy and run your apps from a conversation. - **A connector, nothing to install:** Claude and ChatGPT on the web, desktop and phone. Add the address, sign in once and ask. The code it writes goes into your own repository. - **A local server:** Claude Code, Claude Desktop and GitHub Copilot in VS Code run npx @light-cloud/mcp-server. It can also deploy the folder you are working in. - **Deploy:** Sites and apps from a repository, a local folder or a template, with the build settings read from the code. - **Run:** Environments, logs, scaling, custom domains and environment variables. - **Data:** Postgres and MySQL databases: create, read, change after your yes. - **AI agents:** Create an agent, set its instructions and data, connect it to an app, schedule runs. On Claude’s connector and in the chat. - **Account:** Workspaces, members, plans, and a card through a Stripe link, never the chat. ### [Claude](https://ai.light-cloud.com/mcp/claude) 1. Open Settings → Connectors in Claude and choose Add custom connector. 2. Name it Light Cloud and paste the connector address. Leave the client ID, secret and headers empty. 3. Choose Connect. A Light Cloud page opens and names Claude: sign in and choose Allow. 4. In a chat, turn on Light Cloud under + → Connectors, and ask. Address: `https://mcp.light-cloud.com/mcp` On a Team or Enterprise plan, an owner adds the connector for the organisation first; members then connect it under Settings → Connectors. ### [ChatGPT](https://ai.light-cloud.com/mcp/chatgpt) 1. Open chatgpt.com/plugins, choose +, then Add custom MCP server. 2. Name it Light Cloud and paste the connector address as the public endpoint. 3. Choose OAuth with the scope lightcloud, leave the client ID and secret empty, confirm the notice about custom servers and create it. 4. A Light Cloud page opens and names ChatGPT: sign in and choose Allow. 5. In a chat, pick Light Cloud under +, and ask. Address: `https://mcp.light-cloud.com/chatgpt` ChatGPT gets a smaller set of tools, with no plan or payment tools and no AI agent tools: change those in the console or from Claude. ### [Claude Code](https://ai.light-cloud.com/mcp/claude-code) 1. Run the command at the top of this page. It needs Node.js 18 or later. 2. Check it is connected with claude mcp list. 3. Ask Claude Code to deploy. No account yet? You approve a short code and it is created. Terminal: `claude mcp add light-cloud -- npx @light-cloud/mcp-server` Claude Code asks before each tool call. To approve the Light Cloud tools once, add mcp__light-cloud to the allow list under /permissions. ### [Claude Desktop](https://ai.light-cloud.com/mcp/claude-desktop) 1. Download the latest light-cloud-.mcpb from the releases page and open it. Claude Desktop installs the server with everything it needs: no Node.js, no config file. 2. Or add the server to claude_desktop_config.json as shown, and restart Claude Desktop. 3. The Light Cloud tools appear under the tools icon in the chat box. Ask. claude_desktop_config.json: `{ "mcpServers": { "light-cloud": { "command": "npx", "args": ["@light-cloud/mcp-server"] } } }` ### [GitHub Copilot](https://ai.light-cloud.com/mcp/vscode) 1. Create .vscode/mcp.json as shown, save it and choose Start above the server entry. 2. Open Copilot Chat, switch to Agent mode and check the Light Cloud tools are ticked under Configure Tools. 3. Ask. Copilot asks before each tool; Always allow skips the prompt for that tool next time. .vscode/mcp.json: `{ "servers": { "light-cloud": { "type": "stdio", "command": "npx", "args": ["@light-cloud/mcp-server"] } } }` For every workspace at once, run MCP: Add Server from the Command Palette and choose Global. **Where do my credentials go?** Nowhere near the chat. A connector signs in on a Light Cloud page you approve; the local server signs in with a code you approve in any browser. No password or card number passes through the assistant. **Can it create my account?** Yes. The connector’s sign-in page creates one with Google or GitHub; the local server’s connect tool emails you a link and shows a code to approve. New accounts start on Free with no card. **Can it delete things?** Only if you ask. The assistant confirms first, and your client’s permission prompt is a second gate. **How do I turn it off?** In the console, Settings → Security → Agents & CLI turns off every assistant, or one at a time. Connected devices lists each connection to sign out. **What does it cost?** Nothing extra. It is a way in, not a plan: what it creates is billed by your plan like anything else. **Does it work with other MCP clients?** Any client that can launch npx @light-cloud/mcp-server, Cursor included. The docs have the setup for each. ## Reference - [AI on Light Cloud](https://docs.light-cloud.com/ai/overview): the docs for the chat, AI agents, the gateway API and scheduled runs - [MCP server docs](https://docs.light-cloud.com/deploy-with-ai/mcp-server): every client and every tool - [Light Cloud](https://light-cloud.com/llms.txt): the platform, pricing, runtimes and frameworks - [Help](https://help.light-cloud.com/llms.txt): answers to common questions