# Security — Credential vaulting, BYOK, SSO, audit | Elaichi

> How Elaichi keeps AI access to your SaaS stack safe: vaulted AES-256-GCM credentials the AI never sees, hashed show-once tokens, BYOK AWS KMS envelope encryption, enforced SSO with SCIM, TOTP MFA, and an append-only audit trail.

Security posture of the Elaichi MCP control plane.

## Credentials the AI never sees

Third-party credentials live in a dedicated vault service, encrypted at rest with AES-256-GCM. The control plane stores only credential configurations — the shape of what a connector needs — never secrets. Tools execute server-side with credentials resolved at call time; neither end users nor AI models ever see them.

## Tokens

Session, API, MCP, and invite tokens are stored as keyed hashes and shown exactly once at creation. Lost MCP URLs are rotated (old token revoked first), not left dangling. Rate limits apply on both the data and control planes.

## Bring your own key

Organizations on the Black plan can wrap credential encryption with their own AWS KMS customer-managed keys via envelope encryption.

## Identity and access

Enterprise SSO via hand-rolled SAML SP and OIDC RP — no third-party auth vendor in the login path. DNS-verified domains, enforced SSO, JIT provisioning, SCIM v2 user and group sync, group-to-role mappings, and TOTP MFA with single-use recovery codes.

## Governance

Three-layer RBAC, connector/tool restrictions enforced at connect, advertise, and execute time, and an append-only audit log covering every privileged action — including each MCP tool call with tool, connection, status, and duration — forwardable to your Datadog.

## Offboarding

Removing a member runs a preflight: personal connections referenced by shared toolboxes must be transferred or deleted before removal completes, so shared workflows never silently break.

[Product](https://elaichi.ai/product/) · [Contact](mailto:hello@elaichi.ai)