Get started

Get started

Get your first approval

Create a workspace, send a test request from your browser and approve it, all in about two minutes and with no code.

About 2 minutes

Transcript (30 seconds)

promptcompute puts a person in front of risky agent actions.

Sign up with your work email and create a workspace in one click.

Press Send a test request. A sample refund lands in your inbox.

Read what the agent wants and why, then approve it.

Every decision is signed and kept. Now connect your own agent.

Before you start

Steps

  1. Create your account

    Open promptcompute.ai/signup, enter your work email and press Continue. Open the link we email you; it works once and lasts 24 hours.

  2. Create your workspace

    We fill in a name from your email domain. Keep it or type your own, then press Create workspace. You get a test and a production environment and you are the owner.

  3. Send a test request

    On Getting started, press Send a test request. It creates a sample refund of $120 in your test environment, so it is free, and opens it for you.

  4. Decide it

    Read what the agent asks for and why, then press Approve. You can also lower the amount with Edit, or Reject with a reason. Approvers get the same card by email.

  5. Connect your agent

    Getting started shows one command for Claude Code and copyable snippets for other MCP clients, TypeScript, Python and curl. Pick one and send a real request. If you prefer a terminal, one command with a test key from API keys does the whole tour.

    npx promptcompute init --key pc_test_...

Use cases

See it before you wire anything

The situation
You want to know what an approval looks like before you change any agent code.
What to do
Send the test request and approve it from the inbox and from the email.
What happens
You see the card, the decision and the signed ledger entry in two minutes.

Show a teammate

The situation
A security lead asks what approvers will actually see.
What to do
Invite them under Members, send another test request, and let them decide it.
What happens
They decide it themselves and see that nothing runs until someone approves.

What you will see

Getting started shows the first two steps done, and the inbox shows your test request as approved with a ledger entry.

It worked when

Example output: The test request after approval

Refund $120 on order SAMPLE-1001
medium risk . rule default . agent getting-started sample
Approved by Dana R. at 10:44
History: created, approved  (led_01j...)

If something goes wrong

The sign-in email never arrives
Check spam, then ask again from the sign-up page. Links from older emails stop working once a new one is sent.
The Approve button asks you to confirm it is you
That only happens on high or critical risk requests. Add a passkey or authenticator app under My security, or use the test request, which is medium risk.
You do not see Send a test request
It only shows until the first request exists. Use Open your first request, or send another from API keys with the quickstart command.
Connect Claude Code and other MCP clientsLet Claude Code, Cursor or any MCP client ask for approval before it acts, with one command and a sign-in, no API key.3 minutesSend requests from codeAsk for approval from TypeScript, Python or plain HTTP, wait for the decision, and run only the approved arguments.5 minutesDecide requestsApprove, edit, reject or ask a question from the inbox, email, Slack or Teams, and confirm it is you on high-risk requests.2 minutes