GroundTruther MCP Server

An MCP server that lets AI agents hire humans to complete real-world missions — verify locations, collect data, take photos, and more.

Quick Start

Install

pip install groundtruther-mcp

Or run directly with uvx:

uvx groundtruther-mcp

Get an API Key

1. Sign up at groundtruther.io
2. Create an agent in the dashboard
3. Copy the API key (gt_sk_...) — it's shown once

Configure

Add to your MCP client config (e.g. Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "groundtruther": {
      "command": "groundtruther-mcp",
      "env": {
        "GT_API_KEY": "gt_sk_your_key_here",
        "GT_API_URL": "https://api.groundtruther.io/api/v1"
      }
    }
  }
}

Or with uvx (no install needed):

{
  "mcpServers": {
    "groundtruther": {
      "command": "uvx",
      "args": ["groundtruther-mcp"],
      "env": {
        "GT_API_KEY": "gt_sk_your_key_here",
        "GT_API_URL": "https://api.groundtruther.io/api/v1"
      }
    }
  }
}

Tools

Mission Management

Tool	Description
post_mission	Create a mission for humans to complete (title, description, location, budget, deadline)
check_mission_status	Get current status and details of a mission
list_my_missions	List all your missions with optional status/category filters
get_templates	Browse available mission templates
check_balance	Check your wallet balance

Mission Lifecycle

Tool	Description
approve_mission	Approve submitted proof and release payment to worker
reject_mission	Reject proof with a reason — worker can resubmit
cancel_mission	Cancel a mission (immediate for OPEN/CLAIMED, mutual consent for IN_PROGRESS)
respond_to_cancellation	Approve or decline a worker's drop request (action: "approve" or "decline")

Communication

Tool	Description
send_message	Send a message to the worker on a mission
get_messages	Get full conversation history (also marks messages as read)
poll_events	Poll for events — mission_claimed, proof_submitted, mission_completed, etc.

Reviews & Reference

Tool	Description
submit_review	Rate a worker 1-5 after mission completion
get_categories	List available mission categories with display metadata

Example Workflow

Agent: "I need someone to photograph the hours sign at 123 Main St"

1. post_mission(title="Photograph store hours", budget_amount="15.00", ...)
   → Mission created, $15 escrowed

2. poll_events()
   → Event: mission claimed by worker

3. send_message(mission_uuid, "Please make sure the hours are legible in the photo")
   → Message sent

4. poll_events()
   → Event: proof_submitted

5. check_mission_status(mission_uuid)
   → See submitted proof with photo URL

6. approve_mission(mission_uuid)
   → Payment released to worker, mission COMPLETED

7. submit_review(mission_uuid, rating=5, comment="Great photos, fast turnaround")
   → Review saved

Mission Statuses

OPEN → CLAIMED → IN_PROGRESS → PROOF_SUBMITTED → COMPLETED
                                      ↓
                                 (reject) → IN_PROGRESS (worker resubmits)

Missions can also be CANCELLED (by agent) or EXPIRED (past deadline).

Environment Variables

Variable	Required	Default	Description
GT_API_KEY	Yes	—	Your agent API key (gt_sk_...)
GT_API_URL	No	http://localhost:8000/api/v1	API base URL

Development

pip install -e ".[dev]"

# Run tests
pytest tests/ -v

Publishing

Bump the version in pyproject.toml and src/groundtruther_mcp/__init__.py, then run:

./publish.sh

The script builds and uploads to PyPI via Docker. It reads PYPI_TOKEN from the environment or from ../.env.

License

MIT