OpenClaw Integration

Add policy evaluation, human approval, and audit trails to any OpenClaw MCP server tool.

Overview

OpenClaw agents use MCP servers for tool access. SidClaw's governance proxy sits between OpenClaw and your MCP servers, evaluating every tool call against your security policies before forwarding to the real server.

Quick Start

1. Install the SidClaw skill

openclaw skills install sidclaw-governance

2. Configure the governance proxy

Replace your existing MCP server config in ~/.openclaw/openclaw.json:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "@sidclaw/sdk", "mcp-proxy"],
      "env": {
        "SIDCLAW_API_KEY": "ai_your_key_here",
        "SIDCLAW_AGENT_ID": "your-agent-id",
        "SIDCLAW_UPSTREAM_CMD": "npx",
        "SIDCLAW_UPSTREAM_ARGS": "-y,@modelcontextprotocol/server-postgres,postgresql://localhost/mydb"
      }
    }
  }
}

Every tool call is now evaluated against your SidClaw policies.

How It Works

OpenClaw starts the SidClaw proxy as an MCP server
The proxy starts your real MCP server as a subprocess
When the agent calls a tool, the proxy intercepts the call
SidClaw evaluates the action against your policies
If allowed, the call is forwarded to the real server
If approval is required, the agent is told to wait for human review
If denied, the agent receives a policy denial error

Configuration

Variable	Required	Description
`SIDCLAW_API_KEY`	Yes	Your SidClaw API key
`SIDCLAW_AGENT_ID`	Yes	Agent ID from the SidClaw dashboard
`SIDCLAW_API_URL`	No	API URL (default: https://api.sidclaw.com)
`SIDCLAW_UPSTREAM_CMD`	Yes	Command to start upstream MCP server
`SIDCLAW_UPSTREAM_ARGS`	No	Comma-separated args for upstream
`SIDCLAW_DEFAULT_CLASSIFICATION`	No	Default data classification (default: internal)
`SIDCLAW_APPROVAL_MODE`	No	'error' or 'block' (default: error)
`SIDCLAW_TOOL_MAPPINGS`	No	JSON tool-specific overrides

Tool Mappings

For fine-grained control over individual tools:

[
  {"toolName": "query", "data_classification": "confidential", "operation": "database_query"},
  {"toolName": "list_tables", "skip_governance": true},
  {"toolName": "drop_*", "data_classification": "restricted"}
]

Set via SIDCLAW_TOOL_MAPPINGS environment variable.

Security Context

The ClawHavoc campaign discovered 1,184 malicious skills on ClawHub. SidClaw adds the missing security layer:

Policy-based evaluation of every tool call
Human approval for high-risk actions
Complete audit trail
Tamper-proof integrity hashing

OpenClaw Integration

On this page