humanrail-mcp-server

Route tasks requiring human judgment to a vetted worker pool — directly from any AI agent.

When your AI agent hits something it can't handle — content moderation, refund decisions, subjective quality assessments, data verification — HumanRail routes it to a human worker, verifies the result, pays the worker via Lightning Network, and returns structured output.

Think "Stripe for human judgment."

pip install humanrail-mcp-server

Or run directly:

uvx humanrail-mcp-server

Add to your Claude Code config (~/.claude.json):

{
  "mcpServers": {
    "humanrail": {
      "command": "uvx",
      "args": ["humanrail-mcp-server"],
      "env": {
        "HUMANRAIL_API_KEY": "ek_live_your_key_here"
      }
    }
  }
}

Or for Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "humanrail": {
      "command": "uvx",
      "args": ["humanrail-mcp-server"],
      "env": {
        "HUMANRAIL_API_KEY": "ek_live_your_key_here"
      }
    }
  }
}

Sign up at humanrail.dev to get your API key.

| Tool | Description | |------|-------------| | create_task | Route a task to a human worker for review/judgment | | get_task | Check the status and result of a task | | wait_for_task | Poll until a task completes (blocking) | | cancel_task | Cancel a pending task | | list_tasks | List tasks with filters (status, type, date range) | | get_usage | View usage stats and billing summary | | health_check | Check if the HumanRail API is reachable |

Once connected, Claude can use HumanRail naturally:

> User: "Review this customer's refund request — order #12345, they say the item arrived damaged." > > Claude: I'll route this to a human reviewer for a refund eligibility decision. > *(calls create_task with task_type="refund_eligibility")* > > The human r