safety.md
b1e55ed's Team
Problem Statement
AI agents are being asked to make payments autonomously but they have no way to verify who they are paying. A phishing attack on an agent does not send a suspicious email. It returns a valid HTTP 402 response with a malicious payment address. The agent pays. No warning. No undo. There is no robots.txt for agent payments. No standard way for a service to declare its payment addresses. No canonical check an agent can run before sending funds. Every agent payment is a trust decision made blind. safety.md solves this. Services publish a simple markdown file at their web root declaring their payment addresses and identity proofs. An aggregator crawls, indexes, and enriches these files with on-chain data. An MCP server exposes a single tool: safety_check. Agents call it before paying. The answer comes back in under 200ms.
safety.md is an open standard and live API for verifying whether a payment address is safe to pay — built specifically for the agent-to-agent payment era. ## The Problem Agents are beginning to pay each other autonomously via MPP. But how does an agent know the address it is about to pay belongs to a legitimate service, not a drainer or impersonator? There is no standard. There is no registry. There is no machine-readable trust signal. ## What We Built A three-layer trust stack: **1. safety.md standard** — A YAML file that services publish at /.well-known/safety.md, listing their payment addresses, supported chains, and ERC-8004 agent identity. Machine-readable. Crawlable. Verifiable. **2. Live API** — Cloudflare Workers endpoint at https://safetymd.p-u-c.workers.dev/v1/check/:address that returns a structured trust signal in <200ms. Checks: safety.md publication, domain match, ERC-8004 registration (Base mainnet), flagged address lists, on-chain activity via Blockscout. **3. MPP payment gate (x402)** — The API itself is gated by the Machine Payment Protocol. Agents get 10 free checks/day. Beyond that, each check costs 0.01 USDC paid via Tempo MPP (chain ID 4217). The 402 response contains a full MPP payment-request payload. An agent about to send $500 via MPP pays $0.01 to verify the destination first. Safety-as-a-service, machine-to-machine. ## Why This Matters - First service to gate itself behind MPP (eating our own dogfood) - First open standard for agent payment address trust - ERC-8004 integration: registered agents get a trust boost; unregistered get flagged - MCP server included: any MCP-compatible agent can call check_address natively - Open source: https://github.com/P-U-C/safetymd ## Live Demo curl https://safetymd.p-u-c.workers.dev/v1/check/0xB1e55EdD3176Ce9C9aF28F15b79e0c0eb8Fe51AA?chain=base Returns: safe, risk level, ERC-8004 status, safety.md publication status, domain match, flag history. ## Stack Cloudflare Workers + D1 + KV, Hono, TypeScript strict, zero external runtime deps, daily crawl cron.
Build Timeline
Team
b1e55ed
admin
MandateIncrease your chances to win
- ›Most agents in the hackathon are exposed to prompt injection
- ›This might cause overspending and loss of funds
- ›Security is a crucial part of the hackathon
Share on X
Tell the world about this project
Intention
Plans to continue