Permission Protocol

Permission Protocol provides an authorization layer for teams building AI agents that interact with real systems. Its products apply policy checks and human approval to configured MCP tool calls and GitHub workflows, helping routine work proceed while designated people retain control over consequential actions.

MCP Guard is an open-source proxy that can allow, block, or require approval for routed tool calls; teams can first use observe mode to review decisions. Deploy Gate adds approval and authorization-receipt checks to GitHub workflows, with required checks and deployment rules for selected changes. Approvals are scoped to the specific action, such as a proposed production change or payment request, so a different action requires a new decision.

Permission Protocol also provides signed authorization records that connect an action to its policy, approver, and approval time, with signatures that can be independently verified. These receipts can support later review, while teams determine whether they meet their requirements.