Connect Claude or ChatGPT to Curio
Curio exposes a remote MCP server at https://hey.letscurio.app/mcp. Chat clients that support MCP connectors — claude.ai, Claude Desktop, ChatGPT in developer mode — can connect to it directly and publish artifacts into your Curio account on your behalf, with your explicit consent.
Connect claude.ai or Claude Desktop
- Open Settings → Connectors → Add custom connector.
- Enter the URL
https://hey.letscurio.app/mcpand confirm. - Claude discovers Curio’s OAuth metadata automatically and opens a consent screen in your browser. Sign in to Curio if you aren’t already, review the requested access, and approve.
- Claude redirects back and the connector shows as connected. No API token or manual copy/paste is involved — the whole authorization happens through the browser consent flow.
Connect ChatGPT
- Enable developer mode for connectors (Settings → Connectors → enable developer mode if you haven’t already — MCP connectors are a developer-mode feature in ChatGPT today).
- Settings → Connectors → Add connector, enter
https://hey.letscurio.app/mcp. - Same browser consent flow as above: ChatGPT discovers Curio’s OAuth server, sends you to sign in and approve, then redirects back connected.
What the consent screen shows
Before you click Allow, check what the consent screen tells you: the name of the app requesting the connection, a plain-language description of what it will be able to do (publish artifacts and new versions, and list the ones it can update), which Curio account you’re connecting, and the host the authorization will be sent to — the app’s own domain, for example claude.ai, not hey.letscurio.app. Nothing is authorized until you click Allow.
What the assistant can do
Once connected, the model has three things it can do, and it picks between them based on what you ask for:
- Publish a brand-new artifact and get back its durable review link — used the first time you ask it to publish or share something for review.
- List the artifacts already in your account — used when it needs to find something you published earlier, for example before revising it.
- Publish a new version onto an existing artifact — used when you’re iterating; reviewer comments pinned to the old version carry forward onto the new one automatically.
It cannot read comments yet, and it never sees artifacts or data that belong to other users.
Using Claude Code or scripts
Claude Code and other non-interactive or script clients don’t go through the browser consent flow — they authenticate with a personal API token instead. Create one at Curio → Settings → API & connected apps → API tokens, then add it as a bearer-token MCP connector in your tool of choice. Personal API tokens are shown once at creation, hashed at rest, and can be revoked individually from the same settings page.
Disconnecting
Go to Curio → Settings → API & connected apps. The Connected apps section lists every chat client currently authorized, with its name, when it connected, and when it was last used. Click Disconnect next to any entry to revoke its access immediately — the client’s tokens stop working right away and it will need to go through consent again to reconnect.
FAQ
Do I need an API key?
No — connecting uses OAuth: you sign in and approve in the browser. Personal API tokens exist for CLI tools like Claude Code.
What can the assistant see?
Only your own artifacts: it can publish new work, list what it published, and add versions. It cannot read comments yet, and it never sees other users' data.
How long does the connection last?
Active connections renew automatically. An unused connection expires after at most ~180 days. You can disconnect at any time from Settings.
Does it work with ChatGPT?
Yes — enable developer mode in ChatGPT settings, then add the same connector URL.
Is my work public?
No. Artifacts are private by default or shared by unguessable link — your choice per artifact.