io.github.kaito3636/x402-presign-guard

Official registry Finance and payments

Pre-sign EVM tx/approval safety for AI agents: decode calldata, flag drainers, return GO/HOLD/BLOCK.

What can you do with io.github.kaito3636/x402-presign-guard?

  • unlimited approve / increaseAllowance to an unknown or known-malicious spender
  • setApprovalForAll(true) (classic NFT drainer) to unknown / malicious operators
  • gasless permit / Permit2 approvals (common drainer vector)
  • transfer / transferFrom to known-malicious recipients
  • calls to known-malicious contracts; unknown selectors flagged "simulate before signing"

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

NPM package x402-presign-guard
Claude Code
claude mcp add x402-presign-guard -- npx -y x402-presign-guard@0.1.1
JSON configuration (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "x402-presign-guard": {
            "command": "npx",
            "args": [
                "-y",
                "x402-presign-guard@0.1.1"
            ]
        }
    }
}

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Contents
  1. x402-presign-guard
  2. Why this axis (the honest frame)
  3. What it does (POC — pure code, no keys, no RPC)
  4. Endpoints
  5. Live (2026-10-24)
  6. Use it as an MCP server
  7. The compounding engine (v2, wired but off)
  8. Deploy your own
  9. Distribution (the real lever)

x402-presign-guard

"Should I sign this?" — a pay-per-call pre-sign safety layer for AI trading/wallet agents. An agent POSTs a raw EVM transaction (or an approval) before signing; this decodes the calldata, flags the patterns that drain wallets, and returns GO / HOLD / BLOCK. Settles in USDC via x402 to a self-custody (pseudonymous) wallet.

Why this axis (the honest frame)

10-angle research (2026) on the x402 / agent-payment economy concluded:

  • Real x402 demand is tiny and hyper-concentrated (~$28K/day real commerce; the all-time #1 data seller ≈ $940/mo). Flat/commodity APIs — dates, OCR, raw chain reads — earn ≈ $0. (Confirms why the earlier x402-jp-data fleet made nothing.)
  • The only categories with real crypto-native willingness-to-pay: computed risk/safety verdicts a trading agent must get before it signs — and first-party MEASURED data you generate yourself (no scraping ToS / copyright, un-backfillable time-series).
  • This project is both: a computed pre-sign verdict (not a commodity score card), backed by an accumulating malicious-signature set — the moat that gets bigger the longer it runs.

Honest expectation: $1,000/mo is NOT achievable passively on x402 rails in 2026 — it's above the current top seller. The realistic path is distribution: ship this as an MCP tool that lands in agents' default toolsets (dual-rail: x402 pay-per-call + flat MCP key). x402 is the on-ramp; the MCP install base is the demand. The POC proves the mechanism and starts the compounding asset now.

What it does (POC — pure code, no keys, no RPC)

src/decode.ts statically decodes calldata and catches the real drain vectors:

  • unlimited approve / increaseAllowance to an unknown or known-malicious spender
  • setApprovalForAll(true) (classic NFT drainer) to unknown / malicious operators
  • gasless permit / Permit2 approvals (common drainer vector)
  • transfer / transferFrom to known-malicious recipients
  • calls to known-malicious contracts; unknown selectors flagged "simulate before signing" Known-good routers (Uniswap/Permit2/1inch/Seaport) reduce false positives.

Endpoints

pathpricewhat
GET /guard/token?token=&chain=0.05 USDCpre-TRADE ERC-20 safety for trading agents: malicious-DB + on-chain rug/honeypot red-flags (owner controls, mint/pause/blacklist, renounced?)
GET /guard/tx?to=&data=&value=&chain=0.05 USDCfull pre-sign risk check → GO/HOLD/BLOCK + flags
GET /guard/approval?token=&spender=&amount=freeapproval-specific (unlimited + malicious-spender) — free adoption tier
GET /guard/signaturesfreesize of the accumulating signature set (the moat)
GET /, GET /.well-known/x402freedocs + Bazaar discovery

Live (2026-10-24)

Running in production, paid mode on: https://x402-presign-guard.vercel.app

curl https://x402-presign-guard.vercel.app/.well-known/x402            # discovery (free)
curl "https://x402-presign-guard.vercel.app/guard/tx?to=0x..."         # 402 + payment requirements

A 402 returns the x402 v1 payment requirements (scheme exact, network base, USDC); an x402-capable client pays and retries. Implemented + tested: 25/25 PASS (decode 19 + app 6), tsc clean. Runs on both Cloudflare Workers and Vercel from one codebase.

Use it as an MCP server

Point any MCP client (Claude Desktop, agent frameworks) at the hosted guard — no install build, no npm account needed:

{
  "mcpServers": {
    "presign-guard": {
      "command": "npx",
      "args": ["-y", "x402-presign-guard"]
    }
  }
}

Published on npm: x402-presign-guard (or run straight from source: npx -y github:kaito3636/x402-presign-guard). Exposes two tools — presign_guard_tx and presign_guard_approval — that forward to the hosted guard (GUARD_URL, default the live URL above). The agent calls the guard before it signs.

The compounding engine (v2, wired but off)

wrangler.toml has a commented cron + KV: a scheduled job ingests malicious selectors/addresses from live threat feeds + own honeypot probes, timestamped — so the signature set and the never-backfillable time-series grow every day. That daily accrual is the "掘るほど美味い" asset.

Deploy your own

Vercel (how the live one runs):

npm install
npx vercel deploy --prod                       # free mode first
npx vercel env add PAY_TO production            # Base self-custody wallet 0x… → go paid
npx vercel deploy --prod

Cloudflare Workers (alternative; same code):

npx wrangler login && npx wrangler deploy
npx wrangler secret put PAY_TO && npx wrangler deploy

Distribution (the real lever)

MCP server (this repo, npx github:…) → Official MCP Registry / mcp.so; x402 Bazaar discovery needs a facilitator that implements /discovery/resources (e.g. Coinbase CDP — set FACILITATOR_URL + CDP keys); list on x402scan. The hosted endpoint already serves /.well-known/x402 for crawlers.

Frequently asked questions

What is io.github.kaito3636/x402-presign-guard?

Pre-sign EVM tx/approval safety for AI agents: decode calldata, flag drainers, return GO/HOLD/BLOCK.

How do I install io.github.kaito3636/x402-presign-guard?

It runs on your machine. In Claude Code: claude mcp add x402-presign-guard -- npx -y x402-presign-guard@0.1.1. For other clients, paste the JSON configuration from this page into their MCP settings.

Does io.github.kaito3636/x402-presign-guard need an API key?

Its registry entry lists no required key or secret. Some features may still need an account with the service it connects to.

Is io.github.kaito3636/x402-presign-guard open source?

Its code is public on GitHub, but the repository does not declare a standard license.

Who maintains io.github.kaito3636/x402-presign-guard?

It is published by kaito3636. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.