Testing an AI shopping agent's checkout flow? There's no sandbox for that yet — so I built one

Description and use

acp-sandbox is a hosted mock merchant that implements the ACP checkout API, enabling developers to fully test AI shopping agents’ protocol interactions without real payments.

Problem

AI agents that shop need a sandbox to test the full checkout protocol, not just payment. Existing test modes cover payment but lack a fake merchant to validate session creation, idempotency, order parsing, and retries.

Outcome

Agents can now validate full checkout flows against a realistic mock merchant, ensuring protocol compliance before live deployment.

Workflow

• 1. Obtain a test API key via POST /keys with an email.
• 2. Create a checkout session via POST /checkout_sessions with line_items and currency.
• 3. Retrieve, update, or cancel the session as needed using the session ID.
• 4. Complete the session by POSTing any payment_data to /checkout_sessions/{id}/complete.
• 5. Inspect logs via GET /logs to debug agent interactions.

Evidence

It implements the real checkout_sessions lifecycle from ACP's 2026-04-17 spec: create, retrieve, update, complete, cancel.

Source

Every request/response is logged per API key (GET /logs), so you can see exactly what your agent sent when something doesn't work.

Source