Jimschubert Hi

[](https://github.com/jimschubert/hi/actions/workflows/build.yml) [](https://github.com/jimschubert/hi/blob/main/go.mod) [](https://github.com/jimschubert/hi/blob/main/LICENSE) [](https://goreportcard.com/report/github.com/jimschubert/hi) [](https://github.com/jimschubert/hi/releases/latest)

> A single notification system for every agent.

hi ("human intelligence") is a work-in-progress persistent desktop daemon which provides AI agents (Claude, Qwen, Copilot, and non-alliterative others) a single point of human interaction. As an MCP server, this avoids ending a session where possible which might save you credits/requests.

Here's how it works.

1. Any agent can send a hi_* request to the hi daemon
2. The daemon will display a notification to the user that the agent is awaiting human input
3. The human (you) clicks on the "Show next request" in your system tray
4. hi presents you with a UI for text input, single or multiple choice, confirm, etc. and you respond to the request

Disclaimer

hi is a tool for human-in-the-loop agent interactions. Do not rely on agents for critical decisions without human review.

I'm making this tool available as-is, just please don't use it to generate "slop".

See Anthropic's Framework for Developing Safe and Trustworthy Agents for details on responsible agentic AI.

>[!WARNING] > Be aware that some agents may trigger automatic bans if you're running long-running tasks of many steps. > For instance, see this discussion topic.

Install

go install github.com/jimschubert/hi/cmd/hi@latest

Usage

The hi binary runs the daemon, processes MCP request over stdio, and manages the daemon over IPC.

$ hi --help                       
Usage: hi  [flags]

Human Intelligence — MCP server for human-in-the-loop agent interactions

Flags:
  -h, --help       Show context-sensitive help.
  -v, --version    Print version information.

Commands:
  daemon      Run the hi daemon
  status      Query status of the hi daemon
  ping        Ping the hi daemon
  stdio       Query the daemon via stdio over Unix socket
  shutdown    Shutdown the hi daemon (gracefully)
  test        Submit 1+ test requests to the hi daemon over Unix socket

Run "hi  --help" for more information on a command.

MCP configuration

You can serve MCP over IPC (Unix socket), HTTP, or both. The default is IPC because it's the easies to get started.

IPC (Unix Socket)

This is all you need… the stdio subcommand will automatically start the daemon in IPC mode.

{
  "mcpServers": {
    "human-intelligence": {
      "command": "/path/to/hi",
      "args": [
        "stdio"
      ]
    }
  }
}

HTTP

Assuming you've already started hi daemon with --addr localhost:45678:

{
  "mcpServers": {
    "human-intelligence": {
      "url": "http://localhost:45678/mcp"
    }
  }
}

Example Prompt

Here's an example system prompt to ask your LLM to use hi:

When you need user input, decisions, or confirmations, use the human-intelligence MCP server tools
(get_user_input, get_user_choice, show_confirmation_dialog, etc.)
instead of asking me in chat.

This keeps our conversation clean and provides better interaction.

License

Apache 2.0 - see [LICENSE](LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jimschubert
• Source: jimschubert/hi
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.