# Human Use

> Human Use is an open-source MCP server for human-in-the-loop tasks. MCP server that sends agents' tasks to real people's phones and returns their answers as. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** API, CLI, Self-hosted, Web, iOS, Android
- **Website:** https://humanuse.dev
- **Repository:** https://github.com/human-use/human-use
- **MCP Registry name:** dev.humanuse/human-use
- **Categories:** [human-in-the-loop tasks](https://ai-powered.com/categories/human-in-the-loop-tasks)
- **Listing updated:** 2026-11-19

## About

The README describes Human Use as a way to connect people to an AI agent as MCP tools. An agent calls a tool such as photo, errand or patch, a card is printed on the person's phone, and the result comes back to the agent as JSON. Each person, called a unit, is paired to the runtime with a QR code and can be someone on the lead's team or the lead themselves.

The runtime runs on the lead's computer under Node.js 22.13 or newer and is installed with a single npx command that registers it with Claude Code or Codex. A separate driver model, with no tools, turns the person's replies into structured fields, and the README says that the person's own words are kept on the lead page rather than passed to the agent. Goals can run for long periods, with scheduled wake-ups and evidence levels that grade how a result was verified.

The project is described as early (0.1). It is aimed at developers and leads who want an agent to hand physical or real-world errands to people and collect verified results, and it supports several driver backends including Claude, Codex, Ollama and OpenAI-compatible endpoints.

## Features

- **Phone task cards:** Each tool call prints a card on a paired person's phone and returns the answer as JSON.
- **Evidence-graded results:** Results are rated from runtime records down to the person's own unverified statement.
- **Long-running goals:** Goals keep their clock, budget ledger and evidence across agent sessions and can schedule wake-ups.
- **Multiple units:** One runtime can drive several people, each with their own phone, notebook and mood.
- **Relay or local network:** Phones connect through an encrypted relay from any network, or directly on the same Wi-Fi.
- **Selectable driver backend:** The driver can run on a Claude or Codex subscription, a local Ollama model or an OpenAI-compatible endpoint.

## Good for

- Asking a person to run a physical errand and getting a structured result back
- Collecting a verified photo of a location or item for an agent workflow
- Scheduling recurring check-ins or upkeep tasks for a team member through an agent
- Testing how an agent plans and tracks multi-day goals with human steps

## Install

### Claude Code (Terminal)

```sh
claude mcp add human-use -- npx -y human-use
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "human-use": {
      "command": "npx",
      "args": [
        "-y",
        "human-use"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "human-use": {
      "command": "npx",
      "args": [
        "-y",
        "human-use"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "human-use": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "human-use"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "human-use": {
      "command": "npx",
      "args": [
        "-y",
        "human-use"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "human-use": {
      "command": "npx",
      "args": [
        "-y",
        "human-use"
      ]
    }
  }
}
```

## Tools

- `photo`: Asks a unit to take a photo and returns the result.
- `errand`: Asks a unit to run an errand and returns the outcome.
- `patch`: Listed among the tools in the README example.
- `eat`: Schedules a meal reminder for a unit.

## FAQ

### What do I need to run Human Use?

The README says it needs Node.js 22.13 or newer and an agent host such as Claude Code or Codex, with the runtime running on the lead's computer.

### What license is Human Use under?

The repository and the website both list the MIT license.

### Where does the person's text go?

The README says the person's own words go to the lead page, which the agent cannot read, and the driver turns replies into fields for the agent.

### Can Human Use work without internet access to a relay?

Yes. With the relay set to none, the phone pairs to a direct address on the same Wi-Fi, though push notifications are not available in that mode.

### Is Human Use free?

Yes. Human Use is open source, so you can use and self-host it for free.

### How do I add Human Use to Claude Code?

Run this command in your terminal: claude mcp add human-use -- npx -y human-use

### What are the best alternatives to Human Use?

The closest alternatives to Human Use are hands.txt: a human proxy for AI agents and Japan Hands.

## Where this comes from

- Description: written from the project's README, the project's website and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [hands.txt: a human proxy for AI agents](https://ai-powered.com/mcp/hands-txt-a-human-proxy-for-ai-agents): hands.txt: a human proxy for AI agents is a paid MCP server for human-in-the-loop tasks.
- [Japan Hands](https://ai-powered.com/mcp/japan-hands): Japan Hands is an MCP server for human-in-the-loop tasks.

---
Source: https://ai-powered.com/mcp/human-use
