Sim MCP Js

sim-mcp

LLM工具、脚本和自动化的最小MCP兼容框架。

______________________________________________________________________

安装

npm install sim-mcp
npm install -D typescript tsx
npm install -g @modelcontextprotocol/inspector   # (Optional) For MCP stdio testing

______________________________________________________________________

用法

import { toolsJson, ChatAssistant, isMCP } from "sim-mcp";

/** Input for weather info for a city */
export interface WeatherInput {
  /** Name of the city */
  name: string;
}

/**
 * Get weather info for a city.
 * @param wi Weather input
 * @returns Weather description string
 */
export async function Weather(wi: WeatherInput) {
  return `The weather in ${wi.name} is sunny with 75°F.`;
}

/** Input for greeting */
export interface GreetInput {
  /** Name of the person to greet */
  name: string;
}

/**
 * Return a friendly greeting for a user.
 * @param gi Greeting input
 * @returns Greeting string
 */
export function Greet(gi: GreetInput) {
  return `Hello, ${gi.name}!`;
}

/** No input for random city generation */
export interface RandomCityInput {}

/**
 * Generate a random city name using the LLM.
 * @returns City name string
 */
export async function RandomCity(_: RandomCityInput) {
  const bot = new ChatAssistant();
  const res = await bot.solo("Suggest a random city name.");
  return res.text || "Unknown City";
}

const allTools = toolsJson([Weather, Greet, RandomCity]);
const greetTool = toolsJson([Greet]);

if (!isMCP()) {
  (async () => {
    const bot = new ChatAssistant({
      instructions: "Concise and helpful.",
      tools: allTools,
    });

    // PROMPT: User-driven Q&A loop
    let res;
    do {
      res = await bot.prompt("Ask about weather, greeting, or type 'exit':");
      if (res.text) console.log("Assistant:", res.text);
    } while (res.type !== "exit");

    // CHAINING: Use RandomCity, then Weather
    const city = await bot.decide("Pick a random city.");
    if (city.error || !city.result) {
      console.log("Error getting random city.");
      return;
    }
    const weather = await bot.decide(`What's the weather in ${city.result}?`);
    if (weather.error) {
      console.log("Error getting weather.");
      return;
    }
    console.log(`Random city: ${city.result}\nWeather: ${weather.text}`);

    // DISCUSS: LLM-only chat (no tools)
    const chatRes = await bot.discuss(
      "Tell me something interesting about the world's capitals."
    );
    console.log("Discuss:", chatRes.text);

    // DECIDE with custom tool set: only greet
    const greetRes = await bot.decide("Say hello to Alex.", {
      tools: greetTool,
    });
    console.log("Decide (greet only):", greetRes.text);
  })();
}

______________________________________________________________________

辅助方法

• 所有方法都接受以下选项 { model, instructions, tools }.

______________________________________________________________________

API

• toolsJson(tools: Function[])
• ChatAssistant({ model?, instructions?, tools? })

- .prompt(message, options?) - .decide(message, options?) - .chat(message, options?) - .solo(message, options?) - .discuss(message, options?)

• isMCP()

_默认导出_: { toolsJson, ChatAssistant, isMCP }

______________________________________________________________________

MCP工具测试

npx @modelcontextprotocol/inspector tsx yourfile.ts

______________________________________________________________________

环境

• 使用 .env 在项目根 OPENAI_API_KEY 以及其他变量。
• .env 是自动加载的。

______________________________________________________________________

关于

作者：丹·怀特黑德()\ 零配置，LLM和MCP的工具优先自动化。

______________________________________________________________________

许可证

麻省理工学院