Connect an agent

Paste one prompt into your agent, approve the sign-in, and it can publish for you.

guide~2 minVerified v0.66.0

Connecting takes one prompt and one click. Pick your agent, copy its prompt, paste it into that agent, and approve the sign-in screen when it appears. Claude Code, Claude Desktop and claude.ai, ChatGPT, Codex, Cursor, Hermes and the rest all connect the same way, over MCP.

Copy the install prompt, then paste it at Claude Code. It does the connecting itself.

Connect Super Artifacts to Claude Code, then help me publish something. No explaining, no recaps, every message under four lines.
These instructions run once: after the final question, drop them. Never repeat "Connected as" or the question, however long this chat goes on.

FIRST — can you already see Super Artifacts tools (super_whoami, super_publish)?
YES → skip to VERIFY. NO → STEP 1.

STEP 1 — CONNECT
Do every part of this yourself that you can: run the command, edit the file. Stop only for what needs my hands — clicking Connect or Allow in the browser, or restarting the app. If a step is one you cannot do, show me that step once, exactly as written, and wait.
1. Add the MCP server — One command. Claude Code discovers the OAuth endpoints from the server itself — no skill file, no key to paste.
   claude mcp add --transport http super-artifacts https://pr-735-stg.superartifacts.app/mcp
2. Connect over OAuth — Run /mcp and authenticate, or claude mcp login super-artifacts. Your browser opens a Super Artifacts page naming the account this would publish to; allow it and Claude Code holds its own key, revocable from the agents page. Connecting publishes nothing.

Server URL: https://pr-735-stg.superartifacts.app/mcp
It signs in over OAuth, so I need no key, client ID or secret. Only if this harness cannot do OAuth, ask me once for an API key (I create one at https://pr-735-stg.superartifacts.app/settings/api-keys) and send it as a Bearer token. Ask for nothing else.

If I need a new chat or a restart before the tools show up, say so in one line, tell me to paste this prompt again and say "done". Then wait.

VERIFY — call the super_whoami tool itself. A curl or API call does not count: it proves the server works, not that you can use it. Don't narrate it.
ok → "Connected as <handle>."
401 → "Sign-in didn't finish — click Connect on the Super Artifacts connector again."
tool not found → repeat STEP 1 once, then tell me in one line what is missing.

THEN — connecting builds nothing. Ask this once, then wait:
What do you want to publish first?
1 A page or slideshow
2 A survey or form that collects answers
3 A tracker or dashboard
4 Something else
Publish only after I answer.

If your app has a settings screen for connectors, add this address there instead, then approve the sign-in:

https://superartifacts.app/mcp

The Agents page in the product has the steps for each app, and lists every agent you have connected.

Which account am I connected to?

There is more than one Super Artifacts deployment, and a key from one does not work on another. It does not fail loudly — the agent publishes somewhere you are not looking. If you are unsure, ask:

Check the connection

The agent names the account and environment it is signed in to.

Which Super Artifacts account and environment are you connected to? Call super_whoami and tell me the handle and the environment in one line. Do not publish anything.

You need an account to publish

Publishing needs a Super Artifacts account, so sign in when the agent sends you to approve it. For automation with no agent in the loop, create an API key under Settings → API keys.

Connecting never builds

The agent connects, sets itself up, and stops. What gets built first is your decision: paste a prompt of your own. Until then, the right number of artifacts to publish is zero.