OpenClaw is an open-source guardrail that intercepts AI agent actions before they execute. It calls APort's verify API to check policy and returns allow/deny.

How do I add OpenClaw to my agent?

Install the OpenClaw package, configure your agent ID and policy pack, and wrap your agent's execution. OpenClaw supports LangChain, LangGraph, CrewAI, and custom integrations.

Does OpenClaw work with MCP?

Yes. OpenClaw can protect MCP tool calls by verifying each tool invocation against your policy pack before the tool executes.

← All quickstarts

OpenClaw Guardrails in 5 Minutes

OpenClaw already gives you sandboxing and tool policy. Add APort when you need per-agent limits, kill switch, and decision-level audit before tools run.

Have a passport from aport.io? Run npx @aporthq/aport-agent-guardrails openclaw <agent_id> to use a hosted passport and skip the wizard.

Why OpenClaw needs guardrails →

Install Guardrails

One command — plugin install plus passport/config wizard

bash

npx @aporthq/aport-agent-guardrails openclaw

Start OpenClaw

Start the gateway with the generated config

bash

openclaw gateway start --config ~/.openclaw/config.yaml

# The plugin automatically checks EVERY tool call
# against your passport before execution

Try a Safe Command

Tell your agent to create a file

bash

# User: "Create a file called test.txt"
# → Agent calls exec.run
# → APort plugin: ✅ ALLOW
# → Tool executes successfully

File created

Try a Dangerous Command

Watch the guardrail block it

bash

# User: "Run rm -rf /"
# → Agent calls exec.run
# → APort plugin: ❌ DENY - Blocked pattern
# → Tool execution prevented

❌ DENY - Blocked pattern: rm -rf

Customize Your Passport

Edit limits, suspend the passport, or switch to hosted mode

bash

# View passport status:
~/.openclaw/.skills/aport-status.sh

# Edit passport (optional):
nano ~/.openclaw/aport/passport.json

🎉

That's it.

Your OpenClaw agent now has a real pre-tool-call authorization layer. OpenClaw handles containment and tool availability; APort adds per-action authorization, limits, and audit.

Quick Questions

Q: Do I need an account?

A: No for local mode. The setup command can create a local passport and enforce policy entirely on your machine. Hosted mode is optional if you want centralized kill switch and audit.

Q: Doesn't OpenClaw already have security controls?

A: Yes. OpenClaw already has sandboxing, tool policy, elevated exec controls, and install-time scanning. APort adds external authorization, per-agent limits, kill switch, and audit on top of those controls.

Q: Can the AI bypass this?

A: No. The plugin enforces policy before the tool runs. Prompt injection can influence the model, but it cannot rewrite the authorization decision made in the hook.

Q: What if I want more than OpenClaw's built-in tool policy?

A: That is where APort helps: identity-scoped limits, hosted or local kill switch, signed audit decisions, and the same policy model across OpenClaw and other frameworks.

Create Free Passport