Human Use is an MCP server (by Rapidata) that connects AI agents to real human intelligence via the Rapidata API - asking humans for free-text responses to a question, classifying or ranking a directory of images, or comparing two texts and picking which is better. A hosted version is also available at chat.rapidata.ai without setting up your own environment. Demonstrated use cases include coming up with a car design, finding the best slogan, naming a function, and ranking different image-generation models against each other.

Use it when an AI agent needs a genuine human judgment call on a subjective question, rather than guessing at an answer itself. For agents running inside Cursor, or if Claude Desktop is your MCP client, the project's own custom Streamlit app is the recommended way to use the server, since Claude Desktop had client-side issues with other MCP clients. GitHub stars, documentation, and a Twitter/X account are also linked from the project for anyone wanting to follow its development.

Four tools: get_free_text_responses (ask humans for short free-text answers to a question - useful for open-ended feedback like naming or slogan ideas), get_human_image_classification (ask humans to classify every image in a given directory), get_human_image_ranking (ask humans to rank the images in a directory against each other, useful for comparing outputs from different image-generation models), and get_human_text_comparison (ask humans to compare two pieces of text and select which one is better).

For Cursor, add to ~/.cursor/mcp.json:

{
    "mcpServers": {
        "human-use": {
            "command": "uv",
            "args": [
                "--directory",
                "YOUR_ABSOLUTE_PATH_HERE",
                "run",
                "rapidata_human_api.py"
            ]
        }
    }
}

For the Streamlit app: clone the repository (git clone https://github.com/RapidataAI/human-use.git), install uv if needed (curl -LsSf https://astral.sh/uv/install.sh | sh on macOS/Linux, or the PowerShell equivalent on Windows), copy .env.example to .env with your own credentials/settings (using absolute paths), create/activate a virtual environment (uv venv), install dependencies (uv sync), then run streamlit run app.py. If streamlit/python paths mismatch, run python -m streamlit run app.py instead; if uv isn't found, close all terminals/editors and reopen before retrying.

Developers building AI agents that need real human judgment on subjective tasks - design choices, slogans, naming, or ranking model outputs - rather than relying purely on the model's own guess. Questions or setup help can be directed to the Rapidata team directly by email, and full documentation is maintained separately at docs.rapidata.ai.