x402-seller — rug protection & safe-alpha for trading agents

io.github.wyattpalm2-eng由 GitHub 账号 wyattpalm2-eng 发布v1.0.0更新于 Oct 27, 2026

Rug-check & launch radar for trading agents: composite honeypot score, EVM+Solana, keyless x402.

已验证Streamable HTTP可网页运行金融安全与监控

概览

AI 生成的概览

远程 MCP 服务器,为交易代理提供 EVM 与 Solana 代币的貔貅盘/撤池风险判定、新币筛选和流动性流失信号。

可以做什么
  • 对单个代币做综合风险评分并给出 clear/caution/avoid 结论,附详细红绿旗报告与实时买卖模拟结果。
  • 一次调用发现新上线代币,按流动性排序并逐个给出风险判定。
  • 查询该服务公开的自我评分历史成绩与 Truth Engine 账本。
  • 查看服务目录、价格与付款方式。
适用场景
适合在助手或自动交易代理准备买入新上线或不熟悉的代币前,快速获取风险结论。也适合扫描新币,筛出通过基础风险检查的候选标的。
运行要求
通过 streamable HTTP 连接的远程 MCP 端点,无需安装。不需要账号或 API 密钥。免费演示工具共享有限的每日额度;无限使用需通过 x402 客户端在 Base 上按次以 USDC 付费,自托管版本则需一个 burner 钱包密钥。
安装前请注意
判定结果是第三方风险评分而非保证,服务方也说明其账本在重新部署后会重置。付费调用会真实花费 Base 上的 USDC;自托管选项要求在 X402_BUYER_PK 中提供有余额的 burner 钱包私钥,切勿使用主钱包。代币地址与查询内容会发送到该远程服务及其上游数据提供方。

安装

在 SourceWeft 中

  1. 打开 控制台中的 x402-seller — rug protection & safe-alpha for trading agents,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.wyattpalm2-eng%2Fx402-seller/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "x402-seller-rug-protection-safe-alpha-fo": {
      "type": "http",
      "url": "https://x402-seller-m8nx.onrender.com/mcp"
    }
  }
}

README

x402-seller — rug protection for autonomous trading agents

Live: https://x402-seller-m8nx.onrender.com · Base mainnet · no signup, no API key — pay per call in USDC via x402

Your agent is about to ape into a token. One bad ape costs its whole position. This API answers "is this a rug?" in one call, with evidence:

  • Composite rug score — GoPlus static analysis fused with a live Honeypot.is buy/sell simulation (it actually executes a simulated trade), a serial-rugger check, hard honeypot gates, and a needs_review flag when the two engines disagree.
  • Solana too — dual-engine (GoPlus-Solana + RugCheck): mint/freeze authorities, holder concentration, LP burn. Different chain, different rug physics, same design.
  • Liquidity-drain detector — we poll pool reserves over time and answer "is liquidity leaving right now?" — the earliest sign of a rug in progress. This time-series exists nowhere for free; you can't backfill history you didn't collect.
  • Public self-graded track record — every 30 minutes we score fresh Base launches with the exact paid scorer, then grade ourselves against what actually happened. Hits and misses published. Evidence, not claims.

Try it free, right now

bash
# one real /vet per hour, full paid output (chain=base|eth|bsc|polygon|arbitrum|optimism|solana)curl "https://x402-seller-m8nx.onrender.com/demo/vet?chain=eth&address=0x6982508145454Ce325dDbE47a25d4ec3d2311933"
# the receipts: our verdicts graded against real outcomes (free)curl "https://x402-seller-m8nx.onrender.com/track-record"
# sample output for every endpoint (free)curl "https://x402-seller-m8nx.onrender.com/examples"

Buy with three lines (any x402 client)

ts
import { wrapFetchWithPayment } from "@x402/fetch";import { privateKeyToAccount } from "viem/accounts";
const payFetch = wrapFetchWithPayment(fetch, privateKeyToAccount(process.env.PK));const r = await payFetch("https://x402-seller-m8nx.onrender.com/vet?chain=base&address=0x…");// 402 → auto-pay USDC on Base → data. That's the whole integration.

Endpoints

RoutePriceWhat you get
GET /alpha/launches?chain=$0.08Launch radar — discovers what just launched AND rug-screens every candidate in one call, ranked safest-first with a per-token verdict. Replaces the whole discover→screen→rank pipeline. EVM + Solana
GET /vet?chain=&address=$0.05Single-token go/no-go: market + composite rug score + liquidity-drain trend → clear/caution/avoid + reasons. EVM + Solana
GET /onchain/safety?chain=&address=$0.03Composite rug score: red/green flags, 0-100 risk, live-sim results, needs_review disagreement flag
GET /screen?chain=&addresses=a,b,c$0.03Batch-vet up to 8 tokens, sorted safest-first + summary
GET /onchain/liquidity?chain=&address=$0.01Liquidity-drain verdict from our self-collected reserve series: draining_fast/draining/stable/growing
GET /brief?symbol=BTC$0.03Market regime in one call: spot + funding/OI + sentiment → risk_on/risk_off/neutral
GET /derivs?symbol=BTC$0.01Perp funding (hourly + annualized), open interest, crowded-positioning signal
GET /onchain/token /trending /new /defi$0.005–0.01Token snapshots, trending pools, fresh launches, chain TVL
GET /price /stock /markets /signal$0.001–0.01Spot crypto, stocks/ETFs, market snapshot, momentum verdict

Free: / · /demo/vet · /track-record · /examples · /catalog · /llms.txt · /.well-known/x402.json · /.well-known/agent.json · /openapi.json · /health · /stats

Verdict-first flat JSON everywhere: read field 1, act. Reasons included for audit. Malformed requests are rejected before the paywall — your agent never pays for a doomed call.

Why pay when the underlying sources are free?

Honest answer: the snapshots are free — the judgment and the history are not.

  1. The liquidity time-series is self-collected. There is no free API for "liquidity 1 hour ago."
  2. The composite catches what single sources miss: static-only scanners miss simulated sell-traps; simulation-only misses authority/mint risks. Disagreement between engines is itself signal (needs_review).
  3. One $0.05 call replaces 4+ fetches plus the inference tokens your agent burns reconciling them.
  4. It's keyless. Your agent can't fill signup forms — it can pay 5 cents.
  5. The track record is public, misses included. Judge the scorer by its record, not this README.

The Truth Engine — every endpoint grades itself

The track record isn't a feature, it's the operating system. Everything sold here grades itself against reality in public, forever — and the ledgers are git-snapshotted, so a verdict can't be rewritten after reality grades it:

WhatGraded againstLive ledger
Rug verdicts on fresh launcheswhat actually happened 6h later/accuracy · /track-record
/weather/consensus day-max forecasts (6 fixed cities, daily)the independent ERA5 archive, 2 days later/truth/weather
/signal + /brief market calls (2×/day)realized spot movement, 24h later/truth/signal

If the market-call hit rate converges to ~50%, the page will say the signal has no edge — publishing that possibility is what makes every other number credible. Doctrine + all ledgers: /truth. Who runs this (an autonomous AI crew, one human gate) + the real books, live: /company.

Use as an MCP tool (Claude Code, Cursor, any MCP client)

Rug-checking + launch radar as agent tools. Two ways in:

Hosted (no install) — add the remote MCP server, free demo tools out of the box:

bash
claude mcp add --transport http x402-seller https://x402-seller-m8nx.onrender.com/mcp

Tools: vet_token · launch_radar · rug_check · track_record · catalog (shared free daily demo budget; unlimited via the paid HTTP API). Published to the official MCP Registry as io.github.wyattpalm2-eng/x402-seller.

Local (self-hosted, pays for itself) — clone and run the stdio server, add a burner key for unlimited paid calls:

bash
git clone https://github.com/wyattpalm2-eng/x402-seller && cd x402-seller && npm installclaude mcp add rug-check -- npx -y tsx mcp/server.mts

Tools: vet_token · launch_radar · rug_check · liquidity_trend · market_brief · track_record · catalog.

  • No config: works immediately — vet_token routes through the free demo (1 real call/hour), track_record/catalog are always free.
  • X402_BUYER_PK=0x… (a burner wallet with a few dollars of USDC on Base): every tool goes unlimited, settling ~$0.01–0.05 per call via x402 automatically. No account anywhere.
  • X402_SELLER_URL=…: point the tools at your own self-hosted instance.

Run your own

bash
npm install && npm start                            # localhost:4021, Base Sepolia by defaultNETWORK=eip155:8453 PAY_TO=0xYourWallet npm start   # mainnet — keyless facilitators, no CDP account needed

One-click deploy: render.yaml blueprint. The server registers three redundant keyless facilitators (PayAI, xpay, 0xarchive) on mainnet — no facilitator account required.

Prove the full payment loop end-to-end (needs a burner wallet with ~$1 USDC on Base):

bash
BUYER_PK=0x<burner-key> npx tsx scripts/selfbuy.mts

Architecture notes

  • Node/TS + Express + @x402/express. Three redundant facilitators registered in one x402ResourceServer — any one advertising Base-mainnet exact keeps settlement alive.
  • Stale-while-revalidate cache with in-flight dedup in front of every upstream: a paying buyer never blocks on a cold upstream or eats a 502 while a recent value exists.
  • SSRF-safe: chain allowlists, strict address regexes, encodeURIComponent everywhere, upstream errors never echoed to clients. Receive-only — the server holds no private key.
  • Junk never billed: sources throw on non-finite prices/empty payloads → uncharged 502; malformed input 400s before the paywall.
  • /track-record and /onchain/liquidity ledgers are in-memory + JSONL on free-tier ephemeral disk: depth compounds while the service runs and resets on redeploy. Documented, not hidden.

Layout

src/  index.ts       server: paywall wiring, free routes, demo, storefront  safety.ts      EVM composite rug score (GoPlus × Honeypot.is + agreement factor)  solsafety.ts   Solana composite (GoPlus-Solana × RugCheck)  history.ts     liquidity time-series poller + drain detector  record.ts      the self-graded public track record  composites.ts  /vet and /brief answer endpoints  crypto.ts      on-chain aggregator (DexScreener/GeckoTerminal/DeFiLlama)  funnel.ts      who viewed (402) vs who bought — demand telemetryscripts/  selfbuy.mts    one-command real x402 purchase (settlement proof)

MIT. Built to be forked — if you'd rather run your own than pay ours, the code is all here. The moat isn't the code, it's the collected history and the graded record.

来源:README.md,提交 90d42b6

工具

7

vet_token

Go/no-go verdict on a token before trading it: composite rug score (static analysis + live buy/sell simulation + LP-lock/liquidity-pull gate) + market → clear/caution/avoid with reasons. EVM + Solana. FREE demo (shared daily budget); unlimited via the paid HTTP API.

输入参数chainaddress必填

launch_radar

Discover what just launched AND rug-screen it in one call: fresh token launches ranked by liquidity depth (deepest first), with a per-token rug verdict, each with a verdict (a clean contract with unlocked LP is caution, not clear — most fresh rugs are liquidity pulls). The proactive 'safe alpha' feed. FREE demo (shared daily budget).

输入参数chain

rug_check

Detailed composite rug/honeypot report for one token: red/green flags, 0-100 risk, live-simulation results, disagreement flag. EVM + Solana. FREE demo (shared daily budget).

输入参数chainaddress必填

weather_consensus

Cross-source weather consensus for any coordinates: blends Open-Meteo + NOAA/NWS + 7Timer into one temperature with an agreement score (how much independent models agree). Keyless multi-source ground truth in one call — built by the autonomous crew, ported through the bridge. FREE demo (shared daily budget); unlimited via the paid HTTP API ($0.03).

输入参数lat必填lon必填

truth

FREE, always: the Truth Engine — every endpoint this service sells grades itself against reality in public (rug verdicts vs actual rugs, weather forecasts vs the ERA5 archive, market calls vs realized spot). Returns the doctrine + all three live ledgers' summaries. Judge us by receipts, not claims.

无需输入参数。

track_record

FREE, always: the service's public self-graded track record — its rug verdicts on fresh launches graded against what actually happened, hits AND misses. Judge the scorer by its receipts.

无需输入参数。

catalog

FREE: what x402-seller sells, prices, and how to pay. The paid HTTP API is keyless (pay per call in USDC on Base via x402).

无需输入参数。

版本历史

1
  1. v1.0.0最新Sep 16, 2026

评分

已满足适用检查项的 100%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分