mcpField
R

Reins

本地服务
npmjs.com金融数据
访问官网

关于 Reins

为通过 Hyperliquid 交易的 AI Agent 提供风险限额和审计日志，在模型之外强制执行。

配置

本地服务的配置在各客户端中通用，复制后粘贴到客户端的 MCP 配置文件即可。

mcp.json · 通用配置
{
  "mcpServers": {
    "r2rlabs-reins": {
      "command": "npx",
      "args": [
        "-y",
        "@r2rlabs/reins"
      ],
      "env": {
        "REINS_SYMBOLS": "<REINS_SYMBOLS>",
        "REINS_MAX_POSITION_USD": "<REINS_MAX_POSITION_USD>",
        "REINS_DAILY_LOSS_USD": "<REINS_DAILY_LOSS_USD>",
        "REINS_MAX_LEVERAGE": "<REINS_MAX_LEVERAGE>",
        "REINS_MODE": "<REINS_MODE>",
        "REINS_NETWORK": "<REINS_NETWORK>",
        "REINS_REQUIRE_STOP_LOSS": "<REINS_REQUIRE_STOP_LOSS>",
        "REINS_LOG_FILE": "<REINS_LOG_FILE>",
        "REINS_PAPER_FILE": "<REINS_PAPER_FILE>",
        "REINS_ACCOUNT_ADDRESS": "<REINS_ACCOUNT_ADDRESS>",
        "REINS_PRIVATE_KEY": "<REINS_PRIVATE_KEY>"
      }
    }
  }
}
启动命令npx -y @r2rlabs/reins
需要配置的环境变量
REINS_SYMBOLSMarkets the agent may trade, comma separated, e.g. BTC,ETH. Every other market is refused.
REINS_MAX_POSITION_USDLargest position per market, in USD.
REINS_DAILY_LOSS_USDRealised loss in a UTC day after which new risk is refused.
REINS_MAX_LEVERAGELargest total exposure as a multiple of account value.
REINS_MODEpaper (live prices, simulated fills, nothing signed) or live.
REINS_NETWORKHyperliquid network for prices and, in live mode, orders. Mainnet spends real money.
REINS_REQUIRE_STOP_LOSStrue to refuse any order that adds risk without a stop-loss.
REINS_LOG_FILEDecision log, one JSON line per action with the agent's stated reason. In memory if unset.
REINS_PAPER_FILEWhere the paper account is kept between runs. In memory if unset.
REINS_ACCOUNT_ADDRESSLive mode: the Hyperliquid account traded. Approve Reins' 2 bp builder fee from it first: npx @r2rlabs/reins approve-builder
REINS_PRIVATE_KEY密钥Live mode: an API wallet's key, which can trade but not withdraw. Never the account's own key.

本地服务需要在你的电脑上安装运行环境（如 Node.js 18+）。各客户端的配置文件位置不同，请参考客户端文档。

安装包
npm@r2rlabs/reins0.1.5

工具

还没有获取到工具列表

本地服务需要在你的电脑上运行后才能读取工具，维护者暂未提供工具列表。

详细介绍

维护者还没有提供详细介绍，可以先查看代码仓库中的 README。

评论

支持 Markdown

第一次接触 MCP？

从协议是什么，到在客户端里装上第一个服务，这几条回答覆盖了大多数入门问题。

一个开放协议，规定了 AI 应用如何以统一方式连接外部工具与数据源。客户端负责与模型对话，MCP 服务负责提供工具、资源和提示模板，两者通过标准消息通信，同一个服务可以被不同客户端复用。