OpenClaw Integration

Integrate MeshGuard's governance control plane with OpenClaw agents using the official meshguard-openclaw extension. Enforce policies on every tool invocation with zero code changes.

OpenClaw = Clawdbot

OpenClaw is the open-source rebrand of Clawdbot. Existing Clawdbot users can continue using the Clawdbot integration guide — the SDK-based approach works for both.

Installation

Install the official MeshGuard extension for OpenClaw:

# npm
npm install meshguard-openclaw

# pnpm
pnpm add meshguard-openclaw

# Or via OpenClaw CLI
openclaw plugins install meshguard-openclaw

Quick Start

1. Get MeshGuard Credentials

Sign up at meshguard.app and create an agent to get:

• API Key (msk_xxx)
• Agent ID (agent_xxx)

2. Configure OpenClaw

Add to your openclaw.json:

{
  "plugins": ["meshguard-openclaw"],
  "meshguard": {
    "apiKey": "${MESHGUARD_API_KEY}",
    "agentId": "agent_xxx",
    "mode": "enforce"
  }
}

Or use environment variables:

export MESHGUARD_API_KEY=msk_xxx
export MESHGUARD_AGENT_ID=agent_xxx

3. Define Policies

Create policies in the MeshGuard dashboard or via YAML:

name: openclaw-agent-policy
version: "1.0"
appliesTo:
  agentIds:
    - agent_xxx
rules:
  - effect: allow
    actions:
      - "tool:read"
      - "tool:web_search"
      - "tool:web_fetch"
    description: Allow read operations
    
  - effect: deny
    actions:
      - "tool:exec"
    conditions:
      command_pattern: "rm -rf /*"
    description: Block destructive commands
    alert: critical
    
  - effect: allow
    actions:
      - "tool:*"
    description: Allow all other tools

4. Run Your Agent

That's it! MeshGuard now:

• Evaluates policies before each tool call
• Blocks actions that violate policies (in enforce mode)
• Logs all actions to the audit trail
• Sends alerts for policy violations

Configuration Options

Option	Type	Default	Description
enabled	boolean	true	Enable/disable governance
apiKey	string	required	MeshGuard API key
agentId	string	required	MeshGuard agent ID
gatewayUrl	string	https://dashboard.meshguard.app	MeshGuard gateway URL
mode	string	"enforce"	enforce, audit, or bypass
auditLevel	string	"standard"	minimal, standard, or verbose
cacheTimeoutMs	number	60000	Policy cache duration (ms)
failOpen	boolean	false	Allow actions if MeshGuard is unreachable

Governance Modes

Mode	Behavior
enforce	Block policy violations, log all actions
audit	Log all actions, but don't block violations (shadow mode)
bypass	Disable governance entirely

Production Recommendation

Always start with audit mode to validate your policies before switching to enforce.

How It Works

┌─────────────────────────────────────────────────────────────┐
│                      OpenClaw Agent                          │
│  ┌─────────────────────────────────────────────────────────┐│
│  │                    Tool Invocation                       ││
│  └────────────────────────┬────────────────────────────────┘│
│                           ▼                                  │
│  ┌─────────────────────────────────────────────────────────┐│
│  │              before_tool_call hook                       ││
│  │         ┌─────────────────────────────┐                 ││
│  │         │   MeshGuard Extension       │                 ││
│  │         │   - Evaluate policy         │                 ││
│  │         │   - Check cache             │                 ││
│  │         │   - Block if denied         │                 ││
│  │         └─────────────────────────────┘                 ││
│  └────────────────────────┬────────────────────────────────┘│
│                           ▼                                  │
│  ┌─────────────────────────────────────────────────────────┐│
│  │               Tool Execution (if allowed)               ││
│  └────────────────────────┬────────────────────────────────┘│
│                           ▼                                  │
│  ┌─────────────────────────────────────────────────────────┐│
│  │               after_tool_call hook                       ││
│  │         ┌─────────────────────────────┐                 ││
│  │         │   MeshGuard Extension       │                 ││
│  │         │   - Log to audit trail      │                 ││
│  │         │   - Record duration         │                 ││
│  │         │   - Capture result/error    │                 ││
│  │         └─────────────────────────────┘                 ││
│  └─────────────────────────────────────────────────────────┘│
└─────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                    MeshGuard Cloud                           │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐         │
│  │   Policy    │  │   Audit     │  │  Dashboard  │         │
│  │   Engine    │  │   Store     │  │     UI      │         │
│  └─────────────┘  └─────────────┘  └─────────────┘         │
└─────────────────────────────────────────────────────────────┘

Tool Action Mapping

OpenClaw tools are mapped to MeshGuard actions:

OpenClaw Tool	MeshGuard Action
exec	tool:exec
read / write	tool:read, tool:write
web_search	tool:web_search
web_fetch	tool:web_fetch
browser	tool:browser
message	tool:message
Custom tools	tool:<name>

Policy Examples

Restrict Shell Commands

name: shell-governance
rules:
  - effect: deny
    actions: ["tool:exec"]
    conditions:
      command_pattern: "rm -rf *"
    reason: "Destructive commands blocked"
    alert: critical
    
  - effect: deny
    actions: ["tool:exec"]  
    conditions:
      command_contains:
        - "sudo"
        - "chmod 777"
    reason: "Privileged commands require approval"

Limit External Communication

name: comms-governance
rules:
  - effect: allow
    actions: ["tool:message"]
    conditions:
      channel:
        in: ["slack", "telegram"]
    
  - effect: deny
    actions: ["tool:message"]
    reason: "Messaging to other channels requires approval"

Restrict File Access

name: file-governance
rules:
  - effect: allow
    actions: ["tool:read"]
    conditions:
      path_pattern: "/workspace/*"
      
  - effect: deny
    actions: ["tool:write"]
    conditions:
      path_pattern: "/etc/*"
    reason: "System file modification blocked"

Audit Trail

Every tool invocation is logged to MeshGuard's audit trail:

# View recent audit entries
meshguard audit tail -n 20

# Filter by agent
meshguard audit query --agent my-openclaw-agent

# Export for compliance
meshguard audit export --from 2026-01-01 --format csv

Audit Entry Structure

{
  "timestamp": "2026-02-05T12:30:00Z",
  "agentId": "agent_xxx",
  "agentName": "my-openclaw-agent",
  "action": "tool:exec",
  "resource": "ls -la /workspace",
  "decision": "allow",
  "durationMs": 142,
  "traceId": "trace_abc123",
  "metadata": {
    "tool": "exec",
    "exitCode": 0
  }
}

Alerting

Configure alerts for policy violations:

{
  "meshguard": {
    "alerting": {
      "webhook": {
        "url": "https://your-server.com/alerts",
        "events": ["deny", "anomaly"]
      },
      "email": {
        "to": ["security@yourcompany.com"],
        "events": ["critical"]
      }
    }
  }
}

Security Considerations

• Sensitive data redaction: Passwords, tokens, and API keys are automatically redacted from audit logs
• Local caching: Policies are cached locally to minimize latency
• Fail-safe defaults: Actions are blocked if MeshGuard is unreachable (failOpen: false)

Troubleshooting

Extension Not Loading

# Verify installation
npm list meshguard-openclaw

# Check OpenClaw recognizes it
openclaw plugins list

Policy Denials

# Test policy evaluation
meshguard policy test <agent-id> tool:exec --resource "ls -la"

# Check audit log for details
meshguard audit query --agent <name> --decision deny

Connection Issues

# Test gateway connectivity
curl https://dashboard.meshguard.app/health

# Verify API key
meshguard agent list

Related

• Clawdbot Integration — SDK-based integration (works for OpenClaw too)
• Policies — Policy format and syntax
• Audit Logging — Audit log configuration
• Alerting — Alert rules and notifications
• API Reference — Full API documentation