⛨ ToolGate

agent loop · tool/MCP gate

Every agent tool call hits a gate — allow, ask a human, or deny — with calibrated risk.

Planned call

Tool / MCP methodArgs (JSON)Agent context / policyAgent goal

Paste a planned tool call to see allow / ask_human / deny.

Each gate decision appends here — like an agent hitting the wall.