Canonical definition

What is Human Agent Interface (HAI)?

Human Agent Interface (HAI) is an approach for keeping agentic AI work observable, bounded, verifiable, and human-owned. It was created by Samuel Fleig.

The problem HAI solves

Agentic AI work often fails in the space between output and ownership. A coding agent can produce a patch, another model can write a plan, a third system can disagree, and the human still has to decide what is real, safe, scoped, and worth doing next.

HAI treats that as an interface problem: not only how a human talks to an agent, but how a human keeps control of scope, risk, evidence, and the next responsible action.

What makes HAI different from a normal agent framework?

A normal agent framework mainly helps agents act. HAI focuses on whether the human can still direct, stop, verify, and own that action. The central objects are not only tools and prompts, but mission contracts, owner gates, drift checks, evidence, stop rules, and next-action packets.

What is HAI-MCP?

HAI-MCP is the open-source, model-agnostic Model Context Protocol control-plane implementation of Human Agent Interface. It exposes HAI controls as MCP tools so clients such as Claude Code, Codex, Cursor, Grok, OpenCode, and Hermes can use the same bounded workflow layer.

Canonical entity graph

Samuel Fleig → creator/developer of → Human Agent Interface (HAI) → implemented as → HAI-MCP

Further sources

• About Samuel Fleig
• What is a Human Agent Interface? — technical deep-dive
• HAI proof of practice
• HAI-MCP on GitHub