
Rugcheck Ai
io.github.MrWizardlyLoaf由 GitHub 账号 MrWizardlyLoaf 发布v1.2.0更新于 Oct 24, 2026
Solana token safety for AI agents — rug-pull, honeypot & Token-2022 trap detection before you buy.
概览
远程 MCP 服务器,用于筛查 Solana 代币的跑路、蜜罐与 Token-2022 陷阱,并生成未签名的 Jupiter 买入交易。
- 可以做什么
- 扫描 Solana 代币,给出 SAFE/CAUTION/DANGER 结论与 0-100 安全评分,或快速的是/否判断
- 检查铸造与冻结权限、Token-2022 扩展以及掌控代币的钱包
- 测试可卖出性与往返成本,并查看流动性、成交量、代币年龄和持仓集中度
- 估算跑路概率与紧迫窗口、评估诈骗意图评分,并比较或批量扫描最多 10 个代币
- 为通过筛查的代币构建带 Jito 小费的未签名 Jupiter 兑换交易
- 适用场景
- 当智能体在 Solana 上交易或评估 SPL 与 Token-2022 代币,需要在花钱前快速做链上尽职调查时使用,包括尚未被其他平台收录的新发代币。
- 运行要求
- 远程 Streamable HTTP 端点;托管服务无需安装或 API 密钥。自托管需安装 Python 依赖,并通过 SOLANA_RPC 环境变量提供 Solana RPC 地址。部分客户端需借助 mcp-remote 桥接连接。
安装
在 SourceWeft 中
- 打开 控制台中的 Rugcheck Ai,将其添加到工作区。
- 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.MrWizardlyLoaf%2Frugcheck-ai/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。其他 MCP 客户端
把它添加到你客户端的 mcpServers 配置中。
{
"mcpServers": {
"rugcheck-ai": {
"type": "http",
"url": "https://web-production-58d585.up.railway.app/mcp"
}
}
}README
🛡️ RugCheck AI — On-chain Token Safety for Solana AI Agents
[version] [license] [MCP] [python] [transport] [CI]
A lightweight MCP server that reads a Solana token's mint directly from the chain to screen for the common rug & honeypot traps before your agent trades — active mint/freeze authority and dangerous Token-2022 extensions (permanent delegate, transfer hooks, non-transferable, pausable) — then, for tokens that pass, builds the buy as an unsigned Jupiter transaction carrying a small Jito tip for faster inclusion. You sign it; the server never holds keys.
Quickstart (30 seconds)
No install, no API key. Point your agent at the remote endpoint:
Then ask one question before any buy:
SAFE → clear. CAUTION → read the risks. DANGER → don't buy. That's the whole loop.
See it catch a rug
Three real mainnet tokens, one scan_token call each (scanned 2026-06-17 — re-run to verify, live state changes):
The honeypot is the one that matters: you could buy it and never sell. RugCheck AI flags it before your
agent spends a cent — even though the token is too new to be indexed anywhere else. Even USDC comes back
CAUTION, not SAFE, because the issuer can still freeze your balance — the verdict tells you the truth, not
a marketing label.
Tools
Screening
scan_token— full safety report in one call: authority, Token-2022 traps, honeypot, liquidity & holder concentration → SAFE/CAUTION/DANGER + a 0–100 scoreis_safe— quick yes/no gate: one boolean before you tradeverify_token_safety— on-chain audit: mint/freeze authority + Token-2022 traps + live marketcheck_authorities— mint / freeze authority and extension detectionsimulate_sell— can the token actually be sold? (honeypot check)simulate_trade— full round-trip (buy then sell back): real entry/exit cost & round-trip loss %check_liquidity— DEX liquidity, 24h volume, age, buys/sellsholders_breakdown— top-holder concentration (dump risk)token_age— freshness + real trading activityrug_forecast— heuristic rug ETA: probability + urgency window + factorsscammer_dna— intent score (0–100): how much the token's structure looks like a deliberate scamcheck_deployer— the wallets that hold power over the tokencompare_tokens— rank a basket of tokens safest-firstbatch_scan— scan up to 10 tokens at once, one report each
Execution
execute_safe_swap— re-screens the mint, then builds a Jito-tipped Jupiter swap (unsigned); refuses tokens that scan DANGER
Connect
Remote (Streamable HTTP) — no install, point your agent at:
Listed on the official MCP Registry as
io.github.MrWizardlyLoaf/rugcheck-ai.
Self-host:
Add it to your agent
Cline / Claude Dev (VS Code) — in cline_mcp_settings.json:
Claude Desktop — in claude_desktop_config.json:
Cursor — Settings → MCP → Add → Streamable HTTP, then paste the endpoint URL.
Any MCP client — it's a standard Streamable HTTP MCP server; point your client at the /mcp endpoint and the 15 tools appear.
Why
Most agents trade Solana tokens blind. RugCheck AI calls getAccountInfo on the mint and reads the
authorities and Token-2022 extensions itself, so you get a real verdict on a fresh launch instead of
unknown — and a live mint or freeze authority is flagged before you buy, not after.
Use it when
Your agent needs to answer, before it spends a cent:
- Is this Solana token safe to buy — or is it a rug pull?
- Is this a honeypot — will I actually be able to sell after I buy?
- Does the mint have an active freeze / mint authority that can trap or dilute me?
- Is there a hidden Token-2022 trap (permanent delegate, transfer hook) that can drain me?
- Pre-trade screening / token due-diligence for an autonomous trading agent.
Built for AI trading agents, snipers and bots that buy SPL / Token-2022 tokens and need a fast on-chain rug check before entering — then a screened Jupiter route once a token clears.
FAQ
How do I check if a Solana token is safe to buy?
Call scan_token(mint) — one call returns a SAFE / CAUTION / DANGER verdict covering mint/freeze
authority, Token-2022 traps, honeypot (sellability), liquidity and holder concentration, plus a
0–100 safety score.
How do I detect a honeypot before buying?
simulate_sell(mint) checks whether a live sell route exists — a token with no route is effectively
a honeypot even when nothing on-chain formally blocks selling.
How do I check holder concentration / whale dump risk?
holders_breakdown(mint) reports the largest wallets and what share of supply they control — high
concentration means one holder can crash the price on you.
How do I know if a token is a rug pull?
rug_forecast(mint) gives a heuristic rug probability and urgency window from real signals
(authority, Token-2022 traps, concentration, sell pressure, age). check_authorities and
check_deployer show exactly who holds power over the token.
Does it work on fresh / newly launched tokens?
Yes — it reads the mint directly on-chain (getAccountInfo), so you get a real verdict on a token
too new to be indexed elsewhere. token_age shows freshness and real trading activity.
Does it touch my wallet or sign anything?
No. Screening is read-only; execute_safe_swap only builds an UNSIGNED transaction for you to sign —
the server never holds keys, never signs, never sends.
Is it free? Do I need an API key? Remote server, no install, no API key. Point your agent at the endpoint and call the tools.
Status
v1.2.0 — working, actively developed, CI-tested. Open source, auditable — the screening tools are
read-only (getAccountInfo); execute_safe_swap only builds an unsigned transaction for you to sign.
MIT licensed. Self-hostable. Built for Solana trading agents.
来源:README.md,提交 58266dc
工具
0版本历史
1- v1.2.0最新Sep 16, 2026
评分
已满足适用检查项的 85%
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
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分