Deploy from your AI agent.
Your agent can build, preview and test a site on Thora by itself. A person approves what goes live.
One safe path to live.
Connect.
Add the Thora connector to Claude, ChatGPT or another MCP tool, and sign in.
Work.
Your agent creates or edits a site. Its changes go to a branch, never straight to live.
Test.
Thora builds a private preview with its own database. Where it's switched on for your organization, your agent opens it and reads the build logs.
Approve.
Your agent requests the release. A person reviews the exact build and approves it once.
What your agent does, and what needs you.
Your agent can
- Create a site and read or edit its files
- Build a private preview and test it with the logs
- Set up the site's database and run its changes in the preview
- Attach secrets, Storage, scheduled jobs and email, where available
- Save the live hostname you want
- Request a release or a rollback
A person does
- Approves a release or a rollback in the dashboard. One approval covers the release and its database changes
- Types in each secret on a secure page. The agent never sees the value
- Signs in to open the private preview
- Approves any purchase or billing change
Why it's safe to hand over.
An agent connects with your sign-in, or with a token you create and can revoke. It can only act where your role allows, in the organization you choose.
Text in a site, a repository or a tool result can't give an agent more permission. Publishing and rollback stay human decisions, and an agent can't approve its own request.
Calls have limits. Your agent should follow the returned wait or recovery instructions. If a change has an uncertain result, it must check the recorded state before retrying.
Honest limits.
Thora Cloud is in beta. A real person sets up each new workspace by email, so there is no self-serve checkout yet.
Some tools depend on your plan and on what's switched on for your site. Your agent checks what's available before it acts, and says so when something isn't.
Other MCP tools connect with an access token. A token is read-only unless an administrator picks write permissions, and the ones made in the dashboard expire after 30 days by default.
Keep reading.
Claude is a trademark of Anthropic. ChatGPT is a trademark of OpenAI. Thora Cloud is independent and isn't affiliated with or endorsed by either company.