Permission Protocol provides an authority layer for AI agents, ensuring every agent action generates a portable, tamper‑evident receipt signed by a designated human. It integrates with tools like MCP, CI/CD pipelines, SQL databases, and APIs to automatically clear routine calls while holding consequential actions for human approval, giving teams clear accountability for AI‑originated production changes.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents can execute code changes, database writes, and deployments autonomously, but organizations lack a reliable way to prove which human authorized each specific action, leading to audit gaps, regulatory risk, and incidents where agents act without proper oversight.
Solution
Permission Protocol provides an external authorization layer that intercepts AI agent actions across MCP tool calls, CI/CD pipelines, SQL queries, and API requests. For routine operations the layer allows immediate execution, while consequential actions are held for approval by a designated human signer. Once approved, a cryptographically signed, tamper‑evident receipt is generated and attached to the change, creating an immutable audit record that links the agent, the human approver, and the exact payload. The system integrates with existing development and deployment tools, enforcing policy without requiring changes to the agent code. This approach enables organizations to demonstrate clear accountability for AI‑originated production changes and satisfy compliance requirements such as SOX, FINRA, and FDA audits.
Target Audience
Primary customers are compliance, risk, and audit teams in enterprises that deploy AI agents for code generation, data manipulation, and automated deployments, as well as security leaders responsible for AI governance.
Features
- MCP Guard proxy that inspects every Model Context Protocol tool call, allowing low‑risk calls and escalating high‑impact calls to a human signer
- Deploy Gate that blocks agent‑authored merges and production releases until a signed authority receipt is provided
- Portable, tamper‑evident receipts cryptographically signed by an external, air‑gapped human key
- Automatic routing of routine actions in milliseconds to avoid bottlenecks
- Integration with GitHub checks, CI/CD pipelines, SQL interfaces, and API gateways for seamless policy enforcement
- Full audit trail linking each action to the responsible AI agent and approving human, suitable for regulatory examinations