0.1.0 • Published 7 months ago

@seepine/ai v0.1.0

Weekly downloads
-
License
MIT
Repository
-
Last release
7 months ago

@seepine/ai

帮助你快速构建拥有 MCP 能力的 Agent

一、基本用法

依赖安装

npm i @seepine/ai

创建 Agent

import { ChatOpenAI, Agent } from '@seepine/ai'

const agent = new Agent({
  baseURL: process.env['OPENAI_BASE_URL'],
  apiKey: process.env['OPENAI_API_KEY']!,
  model: process.env['OPENAI_MODEL']!,
  prompts: [
    {
      // 使用提示词构建 agent
      role: 'system',
      content: '你是一个中英文翻译专家,将用户输入的中文翻译成英文,或将用户输入的英文翻译成中文',
    },
  ],
})

agent.chatSync('荣誉').then((res) => {
  // 输出结果 Honor
  console.log(res)
})

流式输出

for await (const chunk of agent.chat(input)) {
  if (chunk.type === 'chat') {
    process.stdout.write(chunk.content)
  } else {
    console.log(chunk)
  }
}

二、使用 MCP

创建 MCP

import { McpClient } from '@seepine/ai'

// 这是获取网页内容和bing搜索的mcp
const fetchMCP = new McpClient({
  name: 'mcp-server-fetch',
  type: 'stdio',
  command: 'npx',
  args: ['-y', '@seepine/mcp-fetch'],
})

创建 Agent

import { ChatOpenAI, Agent, McpClient } from '@seepine/ai'

const agent = new Agent({
  baseURL: process.env['OPENAI_BASE_URL'],
  apiKey: process.env['OPENAI_API_KEY']!,
  model: process.env['OPENAI_MODEL']!,
  // 传给 Agent
  mcps: [fetchMCP],
})

使用

for await (const chunk of agent.chat(input)) {
  if (chunk.type === 'chat') {
    process.stdout.write(chunk.content)
  } else if (chunk.type === 'tool_call_begin') {
    // 开始调用mcp
  } else if (chunk.type === 'tool_call_end') {
    // 调用mcp结束
  }
}
0.1.0

7 months ago

0.0.2

7 months ago

0.0.1

7 months ago