@pepdev/mcp

Deploy apps from any MCP-capable AI coding tool (Claude Code, Cursor, Gemini, Codex, etc.) via the Pep backend API.

Quick Start

npx @pepdev/mcp

Configure your MCP client (e.g. Claude Code .mcp.json):

{
  "mcpServers": {
    "pep": {
      "command": "npx",
      "args": ["-y", "@pepdev/mcp"],
      "env": {
        "PEP_API_KEY": "your-api-key"
      }
    }
  }
}

Or store your API key in ~/.pep/config.json:

{ "api_key": "your-api-key" }

Tools

| Tool | Purpose | Key inputs | |------|---------|------------| | deploy | Build and deploy an app | path, name, databases, env_vars, start, build | | status | Check project state | project_id or name or subdomain | | logs | Fetch recent app logs | project_id or name, lines |

How It Works

The MCP server is a thin client — it packages your project source into a tarball, uploads it to the Pep backend API, and streams deploy progress back via SSE. All heavy lifting (Docker builds, Nixpacks, Fly.io provisioning, database setup) happens on the backend.

You only need Node.js and a Pep API key. No Docker, Nixpacks, or infrastructure tokens required on your machine.

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | PEP_API_KEY | Your Pep API key | Read from ~/.pep/config.json | | PEP_API_URL | Backend API URL | https://pep-web.fly.dev |

License

MIT