Let an AI agent deploy

Launchpad ships an MCP server, so an AI agent such as Claude Code or Cursor can check that your app is safe to launch, then set it up, deploy it and check on it while you watch, without ever seeing a secret.

Add it

It's @axonscape/launchpad-mcp on npm, and com.axonscape/launchpad in the official MCP Registry, for clients that look servers up there. Node.js 22 or newer. In Claude Code:

claude mcp add launchpad -- npx -y @axonscape/launchpad-mcp

In Cursor, with one click: Add Launchpad to Cursor. Cursor shows the server's command and asks before adding it.

By hand, in Cursor (.cursor/mcp.json), Claude Desktop and most other clients:

{
  "mcpServers": {
    "launchpad": { "command": "npx", "args": ["-y", "@axonscape/launchpad-mcp"] }
  }
}

Then ask: "Is my app ready to launch?" The launch check needs no credentials. To deploy, start the server with the same provider credentials you'd give the CLI: in the client's configuration ("env": { "VERCEL_TOKEN": "…" }), or from the shell that starts it.

What an agent can do

Tool	What it does
launchpad_launch_check	The launch check on the app's folder: what to fix before launch, and how
launchpad_init	Create launchpad.json for a repository
launchpad_plan	Show what a deploy would do, and what's missing first
launchpad_up	Create, wire and deploy (safe to re-run)
launchpad_status	What Launchpad manages for this app
launchpad_check	Probe the deployed app's health URLs
launchpad_doctor	Which credentials are set
launchpad_env_names	The names of an environment's variables, never values

What it deliberately cannot do

• Read a secret's value. launchpad_env_names returns names only.
• Set a secret, or link an existing resource: both need a value piped in, which stays with you in the CLI.
• Tear anything down. There's no down tool.

That line matters: an agent reads your code and your prompts, so it's the wrong place to hold keys or the power to delete a database.

Working well with an agent

• Ask for a launch check before anything else, and let the agent fix what it finds.
• Ask for a plan first and read it. It lists everything that would be created.
• Let the agent run up and read the log: it can usually fix what it finds, such as a missing health route.
• Keep the secrets step for yourself, between the plan and the deploy.