Setu: Let AI agents use your apps, with you approving every write

Hi Product Hunt, I'm Aniket, one of the makers of Setu.

We built Setu because every team putting AI agents to work hits the same wall: each app needs its own OAuth flow, token storage, refresh logic and API wrapper, and then you have to trust the agent not to do something you didn't intend.

Setu is the layer in between. Your AI (Claude, ChatGPT, Cursor or your own agent) gets one connection to your apps. Setu handles the sign-in, keeps tokens in an encrypted vault, and runs the tool calls.

Approvals, so you stay in control:
- New accounts ask before writes and deletes; destructive actions always need a yes
- Per-app rules: allow, ask first, or never
- Read-only connections for apps you only want searched
- Email alert when something is waiting for you (approving needs a signed-in click, never a link in an email)
- 24-hour undo for supported actions, and an activity log of every call

Workflows, so the agent can do recurring jobs:
- Chain tools with data passed between steps, loops over lists, conditions and retries
- Start by hand, on a schedule (with time zones) or from a webhook
- A step that needs approval pauses the run, and approving resumes it
- Ready-made recipes you can copy, review and switch on

Also: one MCP URL per user with tool search, so the model never loads the whole catalog, and it can run in your own cloud with your own keys.

Thanks for taking a look!