doorno402 0.4.0

The security layer your agent needs — x402 payment middleware

DoorNo.402

DoorNo.402 is security middleware for x402 payment requests. It intercepts every 402 response before your agent pays, validates it against known attack patterns, and blocks anything suspicious before a single transaction is signed.

The problem

x402 lets servers charge AI agents for resources over HTTP. The spec has no trust model. Any server can lie about the price, inject instructions into the description field to hijack the agent's LLM, redirect payment to an unknown wallet, or drain the wallet through rapid micro-payments. Without a security layer, your agent pays whatever it is told.

Install

pip install doorno402

npm install doorno402

Quickstart — Python

from doorno402 import protect, PaymentBlockedError
from x402.clients.httpx import x402HttpxClient
from eth_account import Account

account = Account.from_key(private_key)
client = protect(x402HttpxClient(account=account))

resp = await client.get("https://api.example.com/data")
# raises PaymentBlockedError if the 402 payload is malicious

Quickstart — TypeScript

import { protect, PaymentBlockedError } from "doorno402";

const safeFetch = protect(fetch);

const resp = await safeFetch("https://api.example.com/data");
// throws PaymentBlockedError if the 402 payload is malicious

What it catches

#	Vulnerability	Attack	Status
01	Price Inflation	Server claims $0.01, charges $5	Covered
02	Unknown Recipient	No ENS, zero trust score	Covered
03	Redirect Hijack	Payment redirected to attacker	Covered
04	Prompt Injection	LLM override in description	Covered
05	Budget Drain	Rapid micro-payments	Covered
06	TLS Downgrade	Payment over HTTP	Covered
07	Fake Delivery	Empty response after payment	Covered

Configuration

protect() accepts three optional parameters:

• daily_budget (float) — max USD per day, blocks when exceeded
• mainnet_rpc_url (str) — Ethereum RPC for ENS trust lookups
• raise_on_block (bool) — raise PaymentBlockedError vs return 403

How it works

DoorNo.402 registers an event hook on your HTTP client. When a server returns a 402 Payment Required response, the hook fires before the payment SDK processes it. The intercepted payload is run through a 7-stage validation pipeline: price inflation check, ENS trust scoring, redirect detection, prompt injection scan, budget enforcement, TLS verification, and delivery validation. If all stages pass, the response is forwarded to the payment SDK for execution. If any stage fails, the response status is mutated to 403 Forbidden and the payment is never signed. When raise_on_block is enabled, a PaymentBlockedError is raised instead.

KeeperHub integration

DoorNo.402 validates. KeeperHub executes. Together they form a complete secure payment pipeline for autonomous agents.

import { interceptAndForward } from "doorno402/mcp";

const safeClient = interceptAndForward(keeperHubClient, {
  dailyBudget: 5.0,
});
await agent.run(safeClient);

When DoorNo.402 approves a payment, forward the validated details to KeeperHub's Direct Execution API for guaranteed on-chain execution with retry logic and audit trail.

Live demo

The repository includes 7 deployed Vercel servers — 6 malicious, 1 honest. Each server targets a different vulnerability from the table above. The CLI runs an autonomous research agent across all 7 servers and produces a results table showing what was blocked and how much was saved.

Site	Vulnerability	URL
CryptoInsider	Price Inflation	cryptoinsider-nine.vercel.app
ChainPulse	Prompt Injection	chainpulse-chi.vercel.app
BlockBrief	Budget Drain	blockbrief-rho.vercel.app
NodeTimes	Unknown Recipient	nodetimes.vercel.app
Web3Daily	TLS Downgrade	web3daily-alpha.vercel.app
ComboAttack	All vulnerabilities	combo-dusky.vercel.app
ChainWatch	Honest server	chainwatch-tan.vercel.app

Run the CLI:

cd demo/cli
pip install -r requirements.txt
python run.py

The CLI prompts for a research topic and runs the agent autonomously across all 7 servers.

Project structure

sdk/python/       Python SDK
sdk/ts/           TypeScript SDK
demo/cli/         Interactive CLI with research agent
demo/servers/     6 malicious + 1 honest Express servers
demo/agent/       Standalone agent scripts
demo/landing/     Landing page source
skills/           Agent skill prompt for doorno402

License

MIT