Doberman

situational

Runtime guard dog for AI coding agents that blocks unsafe tool calls before they execute.

Visit github.com →Add to kit Open in Noemium

ledger entry · observed by @whysanesanders

Sits on the MCP execution path and gives every tool call a PASS, AUTH, or BLOCK verdict, stopping destructive commands, secret exfiltration, and prompt-injection-driven actions. It fails closed and logs redacted decisions. The project is early but already ships CI-proven parity matrices and a plugin API for custom guardrails.

Strengths

PASS/AUTH/BLOCK verdict on every tool call before it executes.
Fail-closed with redacted decision logs.
CI-proven parity matrices and a plugin API for custom guardrails.

Known limitations

Some host integrations are still experimental or manual to wire.
Adaptive behavioral features currently run only on the MCP proxy path, not all host hooks.
Static egress parsing can be evaded by runtime-constructed URLs or non-verb channels.

Use it when

Securing MCP-based coding agents against destructive or exfiltrating calls.
Teams that need an open, auditable guardrail they can self-host.

Skip it when

Your host integration is still experimental or manual to wire.
You need behavioral guards beyond the MCP proxy path.
Runtime-constructed URLs or non-verb channels would evade static egress parsing.

Buy-check

·Price — free, Apache-2.0 open source core; enterprise platform on roadmap as of 2026-08-19. Open the pricing page, not a screenshot.
·Free tier — Free tier exists — burn it first.
·Cancel fallout — What happens to your data/projects on cancel — check the ToS, not the FAQ.
·Seat math — Per-seat pricing multiplies quietly. Count seats before the annual plan.
·Who skips it — Also skipped when: Your host integration is still experimental or manual to wire.; You need behavioral guards beyond the MCP proxy path..