Trust
Security & trust
Everything on this page is already enforced in the product or stated in our Privacy Policy — this is the short version, written for the person doing the security review.
Agents never enter card data
Hard runtime guardrail: the instant an agent reaches a card-number field it stops and reports — it never types card details. Agents are instructed to stop at checkout without completing a purchase. Full checkouts run only on your staging environment, with test cards you provide.
Your data trains nothing
Site content, screenshots, and reports are processed to deliver the service — never used to train AI models, by us or by our AI providers under our commercial API terms.
You confirm what's yours
You confirm you own or are authorized to test each site you add — we act as your technical agent and don't independently verify ownership (see Terms). Nothing is installed on your site: no scripts, no snippet, no DNS record.
Every run is capped
Per-agent step caps, token budgets, and hard timeouts. A run shows its cost ceiling before launch and can't exceed it.
What agents can and cannot do
mocktomer's shoppers browse your site in isolated, short-lived browser containers. Their behavior is bounded by guardrails — the card-entry stop and resource caps below are enforced at runtime, not just prompted:
- No card data, ever. The instant an agent reaches a card-number field it stops at runtime and reports — it never types card details. Agents are also instructed to stop at checkout without completing a purchase. End-to-end purchases run only on staging environments you designate, with test cards you provide.
- You attest ownership. You confirm you own or are authorized to test each site you add; we act as your technical agent and don't independently verify ownership (see Terms §4). Nothing is installed on your site — no scripts, no snippet, no DNS record.
- Hard resource caps. Per-agent step limits, token budgets, and a hard job timeout. Runs show a cost ceiling before launch and auto-stop at it.
- Full audit trail. Every step each agent takes — screenshot plus reasoning — is stored with the run and visible to you. Nothing about a finding is a black box.
Your data
Content from sites you submit (page content, screenshots, journey context) is processed by AI providers such as Anthropic solely to generate your journeys and reports. We do not use your data, your sites' content, or your reports to train AI models, and our AI providers process it under commercial API terms that do not permit provider-side training on our customers' data.
Data is encrypted in transit (TLS) and at rest on Google Cloud Platform. Payments are handled by Stripe — card details never touch our servers. Our infrastructure secrets live in Google Secret Manager; integration credentials you provide (e.g. Jira API tokens) are stored server-side, encrypted at rest, and never readable by clients. Production services run under least-privilege service accounts with no long-lived keys.
Share links & outbound webhooks. Replays are private by default. An editor can mint a public replay link — an unguessable 192-bit token URL showing a read-only replay (screenshots, persona reasoning, scores; never emails, account ids, or billing data), excluded from search indexing, rate-limited, and revocable instantly. Deleting a site revokes its runs' share links automatically. Alert webhooks POST a minimal JSON payload (domain, run id, alert summary, scores, report link — no screenshots, no personal data) to a URL you configure; destinations are validated to public endpoints only.
Subprocessors
| Provider | Purpose | Region |
|---|---|---|
| Google Cloud Platform / Firebase | Cloud infrastructure, hosting, authentication, database, file storage, secret storage | United States (us-central1) |
| Anthropic | AI model inference for journey simulation and report generation — no training on customer data under commercial API terms | United States |
| OpenAI | AI model inference for AI-shopper visibility checks — generic category prompts only, no customer site content or screenshots; no training on customer data under commercial API terms | United States |
| Google (Gemini API) | AI model inference for AI-shopper visibility checks — generic category prompts only, no customer site content or screenshots | United States |
| Stripe | Payment processing and billing — mocktomer never stores your card details | United States |
The full list, including operational tooling, is maintained in our Privacy Policy §5.
International transfers
We operate from the United States. Where GDPR/UK GDPR applies, transfers rely on the European Commission's Standard Contractual Clauses implemented by our providers. Details in Privacy Policy §13.
Billing honesty
You're charged only when an AI customer delivers a verdict. Journeys that fail on our side are on us. No metered overages, no silent charges — running out simply asks you to top up.
API keys & the MCP server
The MCP server lets AI assistants and CI pipelines drive mocktomer with a scoped key — built so a leaked key stays a small problem:
- Keys are SHA-256 hashed at rest and shown exactly once at creation — we cannot re-display one.
- Optional auto-expiry (30 / 90 / 180 days) and instant revocation — a revoked key stops working everywhere within a minute.
- Every key carries a daily run cap (default 25) — a circuit breaker against leaked keys and looping agents. Blocked launches consume nothing.
- Keys are scoped to one workspace, can never reach admin surfaces, and repeated failed auth is rate-limited per IP.
- Claude connectors use OAuth instead of a key: you approve access on a consent screen, the connector holds only short-lived tokens, and you revoke it by signing out of (disconnecting) the connector.
- Runs launched by a key follow every guardrail on this page — same spend rules, same never-touch-real-payment hard stop.
For enterprise & agency reviews
DPA, subprocessor list, and this security overview are ready to send — no NDA required. We're a young company and say so plainly: no SOC 2 report yet; what we offer instead is a small, fully documented surface area and full journey transcripts you can audit yourself. Email support@mocktomer.ai.
Report a vulnerability
Found something? Email support@mocktomer.ai with “Security report” in the subject. We read every report, respond fast, and won't take legal action against good-faith research.