For developers

Bring your own AI. We bring the cloud.

LazyFrog is a platform behind an MCP gateway, with workspaces that run real processes, PostgreSQL, git, routing and secrets. Your AI does the driving, and you can see and take everything.

Start free Tool reference

Connect your AI

A remote MCP server over Streamable HTTP with OAuth 2.1 sign-in. It works with Claude, Claude Desktop, Claude Code, ChatGPT and Codex. Each has its own steps on the Connect page.

Connection

Server URL: https://mcp.lazyfrog.dev/mcp
Auth:       OAuth 2.1 (client ID metadata documents or dynamic registration)
Tools:      grouped by action, with read-only ones marked

Plugin for Claude Code

The LazyFrog connector plus six skills that teach your AI the steps: make a chat prototype real, deploy from a checkout with git push, check a site, fix what's broken without a redesign, go live, and enter LF Games.

Claude Code

/plugin marketplace add lazy-frog-dev/lazyfrog
/plugin install lazyfrog@lazyfrog

The plugin is open source (MIT) on GitHub: github.com/lazy-frog-dev/lazyfrog. The skills are text only; everything they do goes through the connector, signed in as you. A whole deploy from a folder on your machine, step by step: Claude Code hosting.

What a project is

A sandbox per project with its own PostgreSQL 19 per environment, a file store with an S3-compatible API, and real processes: web, API, workers and cron, defined in services.json. A worked example with its real files: Next.js with PostgreSQL.

Preview and live site with separate databases, files and secrets. Publishing builds, checks migrations first, health-checks and switches with no downtime. Rolling back takes seconds.

Routing in front of the app: redirects, rewrites, headers, CORS, password-protected paths, IP and country allow lists, in routing.json.

Your own MCP server runs the same way: a remote server over Streamable HTTP gets an HTTPS address AI apps connect to. MCP server hosting, with a real one built and published in two messages.

services.json

{
  "services": {
    "web": {
      "type": "web",
      "dev": "pnpm dev --port $PORT",
      "build": "pnpm build",
      "start": "pnpm start --port $PORT",
      "port": 3000,
      "healthcheck_path": "/api/health"
    },
    "nightly": {
      "type": "cron",
      "schedule": "0 3 * * *",
      "start": "node scripts/report.js"
    }
  }
}

Your code, in git, runnable anywhere

Every project is a git repository, and starts with a Dockerfile and compose.yml; publishing warns if they go missing.
Commits are authored as you, with the AI as co-author. Commits that contain secrets are refused.
Clone over HTTPS with a personal token, and push to update the preview: the push answers with the address of the preview, or what to fix. Push to GitHub through the GitHub App, or to any host with a deploy key.
Import from Lovable, Bolt, v0, Replit or Vercel. Export everything, both databases included.

Terminal

$ git clone https://git.lazyfrog.dev/my-app.git
$ cd my-app
$ docker compose up --build

Coming from Vercel

vercel.json redirects, rewrites, headers	routing.json (imported automatically)
Serverless and edge functions	Long-running processes: a web service, workers, cron
Postgres from the marketplace	PostgreSQL 19 in every project, per environment
Preview deployments per branch	One live preview per project that follows your AI's work
Environment variables	Secrets per environment, build-time ones flagged
Vercel Cron Jobs	cron services (imported automatically)