Agent-gate – fail-closed agent gate and tamper-evident receipts as an MCP server (github.com)

A new tool called agent-gate has been announced, designed to enhance the accountability and reliability of AI agents by implementing a fail-closed mechanism for their output verification. This MCP server allows AI agents to undergo a rigorous self-check before claiming completion of tasks, combining deterministic checks, an independent review, and a tamper-evident receipt system. Once an agent completes its work, it must verify its output against a checklist and provide evidence for evaluation. If any criteria are not met, such as a lack of logged receipts, the output is automatically marked as failed, ensuring that agents cannot misrepresent their outputs. The significance of agent-gate lies in its potential to reduce silent failures in AI systems, which often accumulate unnoticed and can degrade performance. By ensuring that every decision is recorded in a hash-chained receipt that cannot be altered, the tool fosters transparency and accountability among AI agents. The integration of mandatory human approval for irreversible actions further enhances the reliability of decisions made by the agents. Designed to be lightweight and easy to integrate, agent-gate reflects a commitment to promoting high-quality AI outputs while mitigating risks associated with unverified agent decisions.