SidClaw governance proxy

SidClaw is a helper that sits between your AI assistant and the tools it wants to use. Before the AI does something, SidClaw checks your rules and, if the action looks risky, asks a human to approve or deny it. It is handy for anyone letting an AI assistant touch real systems like email, databases, or servers.

What this MCP server does

You connect SidClaw to your AI assistant the same way you connect any other MCP server. When the AI wants to use a tool, the request goes to SidClaw first instead of straight to the tool. SidClaw looks at your policies and decides: allow it, hold it for human approval, or deny it. If approval is needed, a reviewer sees the details and clicks approve or deny. Either way, a record of the decision is saved so you can look back later.

Click to zoom

What you can do with it

• Wrap any MCP server so its tools are checked against your rules
• Pause risky actions and ask a human to approve or deny them
• Block actions that your policies say should never run
• Send approval requests to Slack, Teams, Telegram, or email
• Keep a tamper-evident audit trail of every decision
• Give each AI agent an identity with its own permissions
• Run a local demo dashboard to see how approvals look

Try asking your AI

• “Show me the recent tool calls that were held for approval”
• “What policies are currently active for my agent”
• “Approve the pending email send request from the support agent”
• “List the audit trail for the last hour of agent activity”

What it gives back to you

SidClaw gives back a decision for each tool call: allowed, held for approval, or denied. When something is held, you see a card with the details of who asked, what they want to do, and why. After a decision, you get a record in the audit trail. In the chat, this usually shows up as a short status message or a link to the approval dashboard.

Before you start

Who it's for

Teams and office workers who let AI assistants use real tools and want a human checkpoint before anything risky happens.