ReAct-MCP

# LLM + MCP + RAG ## 目标 - **Augmented LLM** (Chat + MCP + RAG) - 不依赖框架 - LangChain, LlamaIndex, CrewAI, AutoGen - **MCP** - 支持配置多个MCP Serves - **RAG** 极度简化板 - 从知识中检索出有关信息，注入到上下文 - **任务** - 阅读网页 → 整理一份总结 → 保存到文件 - 本地文档 → 查询相关资料 → 注入上下文 ## **The augmented LLM** - [Building Effective Agents](https://www.anthropic.com/engineering/building-effective-agents)  ```mermaid classDiagram class Agent { +init() +close() +invoke(prompt: string) -mcpClients: MCPClient[] -llm: ChatOpenAI -model: string -systemPrompt: string -context: string } class ChatOpenAI { +chat(prompt?: string) +appendToolResult(toolCallId: string, toolOutput: string) -llm: OpenAI -model: string -messages: OpenAI.Chat.ChatCompletionMessageParam[] -tools: Tool[] } class EmbeddingRetriever { +embedDocument(document: string) +embedQuery(query: string) +retrieve(query: string, topK: number) -embeddingModel: string -vectorStore: VectorStore } class MCPClient { +init() +close() +getTools() +callTool(name: string, params: Record<string, any>) -mcp: Client -command: string -args: string[] -transport: StdioClientTransport -tools: Tool[] } class VectorStore { +addEmbedding(embedding: number[], document: string) +search(queryEmbedding: number[], topK: number) -vectorStore: VectorStoreItem[] } class VectorStoreItem { -embedding: number[] -document: string } Agent --> MCPClient : uses Agent --> ChatOpenAI : interacts with ChatOpenAI --> ToolCall : manages EmbeddingRetriever --> VectorStore : uses VectorStore --> VectorStoreItem : contains ``` ## **依赖** ```bash git clone git@github.com:KelvinQiu802/ts-node-esm-template.git pnpm install pnpm add dotenv openai @modelcontextprotocol/sdk chalk** ``` ## LLM - [OpenAI API](https://platform.openai.com/docs/api-reference/chat) ## MCP - [MCP 架构](https://modelcontextprotocol.io/docs/concepts/architecture) - [MCP Client](https://modelcontextprotocol.io/quickstart/client) - [Fetch MCP](https://github.com/modelcontextprotocol/servers/tree/main/src/fetch) - [Filesystem MCP](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) ## RAG - [Retrieval Augmented Generation](https://scriv.ai/guides/retrieval-augmented-generation-overview/) - 译文: https://www.yuque.com/serviceup/misc/cn-retrieval-augmented-generation-overview - 各种Loaders: https://python.langchain.com/docs/integrations/document_loaders/ - [硅基流动](https://cloud.siliconflow.cn/models) - 邀请码： **x771DtAF** - [json数据](https://jsonplaceholder.typicode.com/) ## 向量 - 维度 - 模长 - 点乘 Dot Product - 对应位置元素的积，求和 - 余弦相似度 cos - 1 → 方向完全一致 - 0 → 垂直 - -1 → 完全想法