spawnpoint

spawnpoint plugin for Claude

spawnpoint is the place where your team's AI lives. With this plugin, Claude puts the tools, MCP servers, and apps it builds for you online and hands you a link, in a chat, in Cowork, or in Claude Code. The link is private until you share it, with one person's email or with everyone at your company, and the people you share it with never make an account.

Install

In claude.ai or the Claude desktop app:

1. Open Customize, then Plugins, then Add, then Add marketplace, and enter https://github.com/spawnpoint-inc/spawnpoint-plugin.
2. Install spawnpoint, open its Connectors tab, and click Connect.
3. Sign in to spawnpoint in the browser that opens and click Approve.

In Claude Code, one command sets up Claude Code, Codex, Cursor, Copilot, and many more agents:

curl -fsSL https://getspawnpoint.com/install | bash

Or install just this plugin by hand:

claude plugin marketplace add spawnpoint-inc/spawnpoint-plugin

claude plugin install spawnpoint@spawnpoint

Restart Claude Code, run /mcp, select spawnpoint, choose Authenticate, and approve in the browser. You sign in once; there are no keys to copy.

Then ask for something: "make a page where my running club says who is coming on Saturday, and deploy it with spawnpoint." Claude builds it, deploys it, and gives you the link. Ask it to share the link with a teammate, change the app, or take it down.

What's in it

The spawnpoint connector (the tools Claude uses to deploy) and three skills, in plugin/spawnpoint/skills/:

• `deploy-to-spawnpoint`: picks the runtime (static, Node, Python, or Docker), sends the app, follows the deploy until it is live, and hands back the link. It also covers sharing, databases, file storage, MCP servers, and team apps.
• `security-audit`: checks an app before it ships. It finds leaked API keys and files that should not upload, checks the port and bind address spawnpoint needs, and flags debug mode and admin or upload routes with no login. It fixes what is safe to fix and stops the deploy on what is not. Run it yourself with /spawnpoint:security-audit.
• `open-models`: when an app calls OpenAI or Anthropic, asks once whether to run it on open models through spawnpoint instead (no provider account, no key), picks the model, and makes the edit.

The skills are tested against a mock spawnpoint server before each release, so changes are measured, not guessed.

Not using Claude?

spawnpoint is a remote MCP server. Any MCP client can connect to https://app.getspawnpoint.com/mcp (Streamable HTTP, OAuth sign-in in the browser). It is listed in the official MCP registry as com.getspawnpoint/spawnpoint. Steps for each agent: supported agents.

Update

Run /plugin, open Marketplaces, select spawnpoint, and turn on auto-update. Marketplaces from outside Anthropic have it off by default; after that one toggle, Claude Code picks up new versions on its own.

To update by hand, run these two commands, then restart Claude Code:

claude plugin marketplace update spawnpoint

claude plugin update spawnpoint@spawnpoint

New MCP tools need no plugin update: agents get them from the server.

Links

• Docs: getting started and the MCP server and its tools
• For agents: llms.txt and llms-full.txt
• Security, privacy, and terms
• Questions and security reports: mannan@getspawnpoint.com

This repo is the published plugin only. It is synced from spawnpoint's main repository on every release.