llm-mcp-rag-java

This project is a Java implementation based on [KelvinQiu802/llm-mcp-rag](https://github.com/KelvinQiu802/llm-mcp-rag), used for learning and practicing LLM, MCP, and RAG technologies. The author's demonstration video is located at [LLM+MCP+RAG实战-从0无框架实现极简Agent客户端 OpenAI/智能体/大模型/AI/DeepSeek](https://www.bilibili.com/video/BV1dcRqYuECf/) ## 1 The original project used JavaScript, which is dynamically typed. Converting data structures (such as converting MCP's Tool to OpenAI's Tool) was very concise, and tool call parameters could be passed around casually. However, after switching to a strongly typed language like Java, everything needs to be explicitly defined with types, requiring a lot of mapping code and handling serialization. If the documentation is not clear, it can be particularly troublesome, and it's easy to pass parameters incorrectly, such as passing tool call content in a format that LLM doesn't recognize. ## 2 During testing of the MCP client server, it was found that the program did not exit normally when it ended. The reason was that MCP called a Node service when connecting, and after closing the MCP connection, the Node service did not stop, causing the program to not exit. The solution was to force exit the program (perhaps there is a more correct way?). ## 3 During final testing, it was discovered that embedding prompts always returned a 403 error, possibly due to API limitations, which prevent too many requests from being made in a short period. Removing two pieces of knowledge content and retesting resolved the issue. **<div>Some content</div>** No Chinese characters should appear in the output. All HTML tags are preserved.