What is Agent Security / the MCP Gateway?+
Agent Security is a security gateway that sits between your AI tools (Cursor, Claude, VS Code Copilot) and the MCP servers they connect to. It authenticates the human and the agent, authorizes every tool call against fine-grained policy, logs everything for audit, and supports human-in-the-loop approvals—all without code changes. You adopt it with a single URL switch.
What counts as a monthly active user (MAU)?+
A MAU is any unique identity—human or agentic—that triggers at least one authorization decision in a calendar month. Human delegators (e.g. human|{subject}) and the agents acting on their behalf (e.g. agent|{client_id}) are counted as separate identities.
Do I need to modify my MCP servers or AI agents?+
No. The gateway works as a transparent proxy. Point your MCP clients to the gateway URL instead of the upstream server, and enforcement, consent, and audit happen automatically. No SDKs, no code changes.
How do trust levels work?+
Each tool is classified into Low (read-only), Medium (write), or High (destructive) based on its name. Trust levels are hierarchical—higher levels inherit all permissions from lower ones. During the consent flow, users choose the level to grant their agent, capped by the admin-configured ceiling. On the Enterprise tier, trust levels are fully customizable: define your own risk tiers, reclassify tools, and create org-specific policies beyond the default Low/Medium/High model.
What authentication methods are supported?+
The Consent Service supports email/password, email OTP, passkeys, social providers (Google, GitHub, Microsoft), and enterprise SSO (SAML, OIDC). Admins can configure allowed methods per host, including forced SSO redirect and email-domain restrictions.
How long do sessions last?+
Application sessions persist across MCP client restarts and are stored in Redis. They have a 30-day inactivity timeout (reset on every tool call) and a 90-day hard maximum. Enterprise plans add fully configurable consent windows. Admins can also revoke sessions immediately.
What are the Enterprise-only advanced features?+
Enterprise unlocks agent fingerprinting (behavioral verification), in-session monitoring, shadow agent detection, human-in-the-loop (HITL) approvals for high-risk actions, intent-based access control, permission receipts (tamper-evident audit records), and time-limited consent windows.
Do you offer discounts for startups or open-source projects?+
Yes. We offer special pricing for early-stage startups, non-profits, and open-source maintainers. Reach out via Slack or email to learn more.
