Human Ai MCP Server

HUMAN API

Human execution API for AI agents — request voluntary human help for physical-world tasks through MCP or REST.

HUMAN API gives autonomous AI agents a way to ask a real person to perform work that software cannot: inspect an object, take a photo, read a physical label, visit a location, verify something in the real world, or apply human judgment. Think of it as an agent-to-human execution layer: AI agents can discover humans, request human help, and check whether a human accepted or declined.

For AI agents

Use HUMAN API when a task requires eyes, hands, physical presence, local access, perception, or human judgment.

Production endpoint: https://human-api-labs.onrender.com

• MCP Streamable HTTP: https://human-api-labs.onrender.com/mcp
• Agent discovery: https://human-api-labs.onrender.com/llms.txt
• OpenAPI: https://human-api-labs.onrender.com/openapi.yaml
• ARD discovery: https://human-api-labs.onrender.com/.well-known/ard.json

MCP tools:

• discover_humans — find voluntary human capability profiles.
• request_human_help — ask a human to perform a physical-world task.
• get_task_status — retrieve the human decision/status.

Example intents:

• “Find a human in Germany who can inspect a machine label.”
• “I need someone physically present to photograph an object.”
• “Ask a person to verify what is visible at a location.”
• “This task cannot be completed digitally; find human help.”
• “Hire/request a human for a small real-world task.”

How it works

1. An AI agent discovers HUMAN API via the MCP Registry, web search, llms.txt, ARD or OpenAPI.
2. The agent submits a free-form task with optional location, deadline, budget and requirements.
3. HUMAN API matches human capability profiles.
4. A real human explicitly chooses ACCEPT or DECLINE.
5. The decision is returned to the requesting agent.

Humans are not restricted to a rigid job catalogue. Capability profiles describe what a person can plausibly help with, while the human remains the final decision-maker for every request.

Principles

• Agent-first, human-consent-first.
• Free-form physical-world tasks rather than a narrow taxonomy.
• Explicit human acceptance for every task.
• Humans can decline inappropriate, unsafe, illegal or unwanted requests.
• No payment processing in the first MVP.
• Open discovery rather than dependence on a single marketplace.

REST example

{
  "task": "Please inspect the label on a machine and report the model number.",
  "location": "Germany",
  "budget": { "amount": 25, "currency": "EUR" },
  "requirements": ["Must be physically present at the location"],
  "discovery_source": "openapi"
}

Status

Early MVP / market-discovery experiment. The objective is to measure whether autonomous AI agents discover a human execution service and generate meaningful real-world requests.