Agent Proof Protocol (APP) v2.4

The "TCP/IP of Reality." The Universal Settlement Layer enabling AI Agents to verify, fix, and move assets in the physical world.

---

⚡ What is this?

Agent Proof Protocol (APP) is an open-source standard that connects Autonomous AI Agents to physical labor and verification networks. It solves the "Hallucination of Action" problem by enforcing Proof of Physics (Wi-Fi RTT, Bio-Liveness, Geofencing) before releasing funds.

It acts as a neutral middleware, aggregating execution APIs (Uber Direct, TaskRabbit, Field Nation) into a single, audit-ready standard compatible with:

• Local Agents: OpenClaw, AutoGPT (via CLI)
• Cloud Agents: CrewAI, LangChain, Claude Desktop (via MCP)
• Web Agents: Chrome/Gemini Nano (via WebMCP)

📄 Read the Full Whitepaper (v2.4) for architecture, economic thesis, and security specs.

---

📦 Installation

pip install agent-proof-protocol

---

🚀 Quick Start

1. The "Builder" (CLI / OpenClaw)

Perfect for sovereign agents running locally. Verify reality directly from your terminal.

# Verify if a store is actually open (Ground Truth)
agent-proof verify --location "Starbucks 5th Ave, NYC" --intent "Check line length"

# Output:
# {
#   "status": "VERIFIED",
#   "proof": "rtt_hash_8x99...",
#   "image_url": "https://api.agentcrowds.com/proof/8x99",
#   "metadata": {"line_length": "Short", "open": true}
# }

2. The "Suit" (Python SDK / CrewAI / LangChain)

Integrate physical skills into your existing agent swarm.

from agent_proof import PhysicalTask

# 1. Create a Task (e.g., Fix a Server)
task = PhysicalTask.create(
    primitive="FIX",
    subtype="IT_REPAIR",
    location="Data Center B, Austin TX",
    budget=15000, # $150.00
    verification="RTT" # Require Wi-Fi Proof
)

# 2. Wait for Ground Truth (Blocking or Async)
print(f"Task dispatched! ID: {task.id}")
# The funds are held in Smart Escrow until the worker scans the server's Wi-Fi.

3. The "Cloud" (Model Context Protocol)

Run APP as an MCP Server to give Claude Desktop or remote LLMs physical hands.

npx @agent-proof/mcp-server start

Add this to your claude_desktop_config.json to enable "Physical Tools" in Claude.

---

🛠 Architecture: The Tri-State Model

We do not force you to adapt to us. The Protocol adapts to your Runtime.

Runtime	Interface	Best For
Local (Sovereign)	agent-proof Binary (CLI)	OpenClaw, AutoGPT, Shell Scripts
Cloud (Enterprise)	Model Context Protocol (MCP)	CrewAI, LangChain, Claude, OpenAI
Web (Browser)	WebMCP (navigator.modelContext)	Chrome Agents, Gemini Nano

---

🛡 Security & Privacy

• Zero-Trust Egress: Enterprise users can configure base_url to point to private gateways.
• On-Device Anonymization: Images are processed at the edge. Faces/PII are blurred before transmission.
• Sandbox Mode: Pass sandbox=True to simulate physical completion without dispatching humans or spending crypto.

---

🤝 Contributing

We are building the standard for the Agentic Economy.

• Submit Adapters: Help us connect more labor networks (DoorDash, Rappi, etc.).
• Improve Bridges: Help us build better integrations for LangChain and AutoGen.

See CONTRIBUTING.md for details.

---

📜 License

MIT Open Source.

Created by Roger Kwasi Senam Mensah.