Security built in, not bolted on.
mcpgate sits between your AI clients and your tools. That position demands real controls — here is exactly what we do, and what we don't claim.
ISO/IEC 27001:2022 — aligned, not certified
mcpgate's information security management is built on and aligned with the controls of ISO/IEC 27001:2022 (Annex A) — encryption, access control, logging, secure development, supplier and incident management. We operate these controls today and maintain them as a living management system.
To be precise: mcpgate is not (yet) certified. No accredited certification body has audited this management system. We treat that distinction as a matter of honesty — when we are certified, we will say so, with the certificate.
What we actually do
Encryption
OAuth tokens and sensitive data are encrypted at rest (Google Cloud KMS or AES-256-GCM). All traffic is TLS, with HSTS enforced.
Access & authentication
OpenID Connect / OAuth 2.0 with PKCE, per-service authorization, and an explicit allow-list guest model. Administration is token-gated.
PII & privacy
Personal data is pseudonymized before it ever reaches an LLM and rehydrated only for the actual tool call. Logs carry hashed identifiers, never raw emails or IPs. GDPR data-subject rights are supported.
Auditability
Security-relevant events — sign-ins, tool calls, admin actions — are recorded in an append-only audit log using hashed identifiers, ready to pipe into your own SIEM.
Secure development & supply chain
Every release runs through CI with automated dependency vulnerability scanning (Renovate, pip-audit, Trivy) and static analysis (Bandit); the security test suite must pass before a build ships.
Hosting & data residency
The hosted services run in the EU — Hetzner, Falkenstein, Germany — on infrastructure certified to ISO/IEC 27001:2022 and BSI C5. Self-hosted deployments stay entirely on your infrastructure.
Backup & continuity
Nightly off-server full-disk backups plus daily encrypted database snapshots; encryption keys are escrowed off the server so backups stay recoverable.
Incident response
A documented response process with a coordinated vulnerability-disclosure path and a GDPR breach-notification procedure.
An AI-governance layer, not just a proxy
Because mcpgate sits on the path between your AI clients and your real tools, it is the right place to govern what AI can see and do:
Least-privilege tool access
AI reaches only the services you allow-list — scoped per user and per service, not all-or-nothing.
Human in the loop
Destructive actions require explicit confirmation; hooks can enforce policy checks before a tool call runs.
Data governance
Personal data is pseudonymized before it reaches the model, and oversized responses are capped.
Accountability
Every AI tool call lands in the append-only audit log; the throughput view flags unusual data pulls.
Found a vulnerability?
Report it privately — please don't open a public issue. We acknowledge within two business days and work under coordinated disclosure: details go into the release notes once a fix has shipped.