Sentinelgate

SentinelGate acts as an MCP proxy that sits between your AI agent and the MCP endpoints. It enforces policies for every tool call, shell command, and data access, before the agent ever executes anything. With CEL-based rules, per-identity policies, and an admin UI, you can create deny/allow decisions, audit trails, and quota limits. The built-in policy templates cover common security postures (Safe Coding, Read Only, Full Lockdown, etc.), and the system exposes a single /mcp endpoint for your agent while maintaining separate identities and sessions for auditing and isolation. Use the Admin UI at /admin to manage identities, upstream MCP servers, policies, and to view live audit events, or test policy behavior in the Policy Test section.

Prerequisites:

• A machine with internet access and a supported OS (Linux/macOS/Windows).
• Basic shell access (bash, zsh, or PowerShell).
• Optional: curl or a terminal that can fetch scripts from GitHub.

Installation options:

1. Quick install (recommended):

   • macOS/Linux: curl -sSfL https://raw.githubusercontent.com/Sentinel-Gate/Sentinelgate/main/install.sh | sh
   • Windows PowerShell: irm https://raw.githubusercontent.com/Sentinel-Gate/Sentinelgate/main/install.ps1 | iex After installation, run the server with the recommended command: sentinel-gate start

2. Manual download / build from source:

   • Download the prebuilt binaries from the releases page and add sentinel-gate to your PATH.
   • Or build from source (Go 1.24+): git clone https://github.com/Sentinel-Gate/Sentinelgate.git cd Sentinelgate go build -o sentinel-gate ./cmd/sentinel-gate mv sentinel-gate /usr/local/bin/

3. Run the server: sentinel-gate start

Additional setup (optional):

• Open http://localhost:8080/admin to configure policies, upstreams, and identities.
• The MCP endpoint is http://localhost:8080/mcp. Point your agent to connect there with an API key.

Configuration can be customized via optional YAML (server.http_addr, rate_limit, etc.) or through the Admin UI.

Tips and common issues:

• Default port is 8080. If you run behind a reverse proxy, ensure the proxy forwards /mcp and /admin correctly.
• API keys are required for agent authentication. Generate and distribute per-identity keys from the Admin UI.
• Use the Policy Templates to rapidly apply standard security postures, then customize per-identity policies as needed.
• The built-in audit trail captures identity, decision, timestamp, and arguments; use export to CSV or stream via SSE for integration with SIEM.
• If you need zero-downtime upgrades, run a new binary alongside and switch the upstream endpoint after validating policies.
• For offline installs, you can download a release tarball/zip and run the included sentinel-gate binary without additional dependencies.
• CEL expressions can reference identity_roles, tool names (action_name), and content in arguments; test patterns in the Policy Test area before applying to production.