Skip to main content
Vibecode has an MCP server, so you can use it from the AI tools you already use. Connect it once, sign in to your Vibecode account, and ask your AI for an app. Vibecode’s agent builds the app, tests it, and saves it to your account. You can keep working on it in Vibecode on the web or in the Vibecode iPhone app. The server URL is:

What You Need

  • An AI app that supports MCP servers with sign-in. ChatGPT, Claude, Claude Code, Codex, Cursor, VS Code and Gemini CLI all do.
  • A Vibecode account. You can create one on the sign-in page when you connect. New accounts that sign up with Google or Apple get their first app build free. After that, builds and changes use credits.

Connect Vibecode

ChatGPT and Claude

Coding Tools and Other Apps

1

Add the server

-s user makes Vibecode available in every folder, not just the current one.
2

Sign in

In Claude Code, run /mcp, select vibecode, then select Authenticate. Your browser opens the Vibecode sign-in page. Sign in and select Allow.
New to Claude Code? How to Use Claude Code covers installing it, what it costs, and building apps with it.

What Your AI Can Do

Ask in plain English, for example:
  • “Using Vibecode, build me an iPhone app to track my workouts, with a weekly summary.”
  • “Build me a web app to plan my meals for the week.”
  • “Show me my Vibecode apps.”
  • “Make the buttons in my workout app blue and add a rest timer.”
  • “Why did the last build of my habit tracker fail?”
  • “Deploy my meal planner so my friends can use it.”
A build usually takes about 10 minutes. A new app with a backend (a full-stack app) can take about 30 minutes. Your AI can wait for it and tell you when the app is ready, with links to open it in Vibecode, see it on your phone in the Vibecode iPhone app, or publish it to the App Store. What Vibecode offers depends on the app. Other apps means Claude Code, Codex, Cursor, VS Code, Gemini CLI and any other MCP app. They show no app card, so your AI gives you links instead. Deploying puts an app’s latest saved version online, the same as the Deploy button in Vibecode: a web app’s site, at its vibecode.run address or your own domain, or the server behind a mobile or iPhone app. First, Vibecode saves any changes made in the app’s workspace that aren’t saved yet, such as edits made with run_command, as a new version, and deploys that. Your AI never starts a charge. If deploying would start a trial or a charge, nothing is deployed, and your AI gives you a link where you see the price and confirm. The app deploys as soon as you confirm. On the Free plan, the link goes to the plans instead, since plans include live backends. In ChatGPT, you deploy in Vibecode. Choose Your Deployment Path has the details.

Commands in Coding Tools

Claude Code, Codex, Cursor, VS Code, Gemini CLI and other MCP apps get run_command. ChatGPT and Claude don’t. run_command runs a shell command (with bash) in the app’s workspace, a Linux sandbox in the app’s folder, and returns the output. Use it to run tests, check versions, install a package or fix a file. A command can be up to 20,000 characters. On Free, only run_command counts toward the 50. Reading files and logs doesn’t, and a command that couldn’t run isn’t counted. When 10 or fewer are left, each answer says how many. Once all 50 are used, run_command doesn’t run, and your AI gives you a link to the plans. They reset on Monday. On Plus, Pro and Max, run_command waits about 40 seconds. If the command is still going, the answer says it’s running, gives its command_id, and shows the output so far. Your AI then calls get_command_output with that id. Each call waits up to about 40 seconds and returns the new output, and the exit code once the command ends. A command is stopped after 20 minutes, or once it has printed more than 50 MB. When it ends, anything it started that still prints into its output stops too. To keep a server running, send its output to a file, as in nohup <command> > server.log 2>&1 &. If the workspace restarts, a running command is lost, and your AI has to run it again. run_live_command runs a command on the server of an app you’ve deployed with a backend, in its backend folder. That’s the live app, with real users. It backs up the live database first, and you can restore that backup from the app’s Database tab in Vibecode. A command can be up to 4,000 characters and stops after 45 seconds. On Free, your AI is told it needs a plan. Your app’s keys are hidden in everything these tools return, and read_file doesn’t show .env files.

Your Account and Your Data

  • Your AI only gets access after you sign in and select Allow. The sign-in page names the app asking for access.
  • It can see your Vibecode apps and their code. Your app’s keys stay hidden in files, logs and command output.
  • Builds and changes use your Vibecode credits, the same as building in Vibecode.
  • To learn what Vibecode receives and sends back, see Vibecode in ChatGPT and Claude in our Privacy Policy.

Disconnect

Remove the server in your app’s MCP or connector settings. In the command-line tools:
Your apps stay in your Vibecode account.

Troubleshooting

  • The sign-in page doesn’t open. In Gemini CLI, run /mcp auth vibecode. In VS Code, select Allow, then Open. In Cursor, select Authenticate next to Vibecode. In Codex, run codex mcp login vibecode.
  • Claude Code shows two Vibecode servers. Connectors you add in Claude also appear in Claude Code, as claude.ai Vibecode, with Claude’s tools. The one you add with claude mcp add has all the tools in the table above.
  • A build doesn’t start. Your account may be out of credits. Your AI tells you why, and you can add credits or choose a plan in Vibecode.

Vibecode MCP or Vibecode CLI?

With the MCP, Vibecode’s agent writes the app and your AI directs it. With the Vibecode CLI, your own agent writes the code and uses Vibecode’s sandboxes, databases, sign-in and hosting to run and publish it. For help, visit Support.
Last modified on November 5, 2026