The group.raposa/raposa-mcp Model Context Protocol (MCP) server enables a human approval step for AI agents. It provides one tool that routes an action, context, and risk for a decision by a named person. If no explicit approval is recorded, the result is not treated as consent.

🛠️ Key Features

• Human-in-the-loop approval workflow
• Approval result model: approved: true|false
• “Silence is not consent”: timeout, expiry, or rejection returns approved: false
• Hash-chained audit log with exportable decisions

🚀 Use Cases

• Requiring a named person to approve or reject agent actions
• Enforcing approvals based on provided action, context, and risk
• Maintaining an auditable trail for agent decisions

⚡ Developer Benefits

• Clear decision contract via request_human_approval(action, context, risk)
• Audit log integrity via hash-chained logging
• Auditor-friendly export of decisions
• EU-hosted deployment and DPA availability

⚠️ Limitations

• Approved responses are only true when a named person explicitly presses Approve
• No implied approval from silence; expired or timed-out requests yield approved: false