Documented vulnerabilities (August 2025): sandbox escape, over-privileged tokens.
Security research confirmed.
Open Source Security
Security for AI agents
Open-source RBAC firewall for MCP. Deterministic policies, full audit logging, self-hosted.
The Problem
Why blanket permissions are dangerous
Blanket Access
AI agents currently get unrestricted permissions to your tools. Once they are in, they have keys to the kingdom.
Optional Consent
Current MCP security relies on optional user consent prompts. It’s manual, error-prone, and doesn't scale.
Catastrophic Risk
One mistake = deleted databases, leaked data, or arbitrary code execution. Documented vulnerabilities exist.
🚨
How Sentinel Gate Works
Sentinel Gate sits between your agents and MCP servers.
Deterministic RBAC
Rules, not AI intent guessing. Define strictly what each role can do with deterministic policies that never hallucinate.
Tool-Level Control
Granular permissions per role. Grant access to specific tools (e.g., `read_only`) while blocking dangerous ones (e.g., `delete_db`).
Full Audit Logging
Every decision is logged for compliance. Track exactly which agent accessed which tool and when.
Agent
Sentinel Gate
RBAC + Audit
MCP Server
Intercepts every call. Enforces policy. Logs the result.
Open Source
AGPL-3.0
Self-Hosted
Built with Go
Get Started
Secure your AI infrastructure today.
Need Enterprise features? SSO, SLA, dedicated support.
Contact us →