当前位置:首页 > 报告详情

4823 - 利用 MCP 和 Open Liberty 在新的智能体 AI 世界中赋能企业级 Java 工作负载.pdf

上传人: 竿*** 编号:982949 2025-11-29 33页 1.56MB

1、IBM TechXchange 2025 conference 4823Liberty Model Context Protocol(MCP)ServerDon Bourne,IBM Application RuntimesAI&Technology Incubation ArchitectIBM TechXchange|2025 IBM Corporation1#IBMTechXchange2Your AI App Wants to Use Your Services!3You could tell your AI app to use them as is but connecting y

2、our AI app to a long,static list of services is brittle.and each service may have lots of options,which can be challenging for an LLM.and the services may not be documented in a way an LLM can understand.4 and some services return too much information5 and some services return too much information6

3、and its not always clear which combination of services to use,in which order,to achieve the desired result.7Model Context ProtocolOpen Source Standard for interaction between AI apps and supporting toolsUsing MCP,an AI enabled application can access:Tools(functions the AI can call)Resources(data/fil

4、es)PromptsAn MCP server can use different transports:Stdio*:Client launches server as a subprocess and communicates via standard input/outputStreamable HTTP:Client connects to server over network*not included in mcpServer-1.0 beta feature8MCP Tool Best Practices Describe your tools with an LLM in mi

5、ndUse clear/intuitive tool names,clear explanations of parameters,and provide guidance on when to call the tool.Be tolerant of variances in inputLLMs dont always follow instructions precisely Return only the data neededLLM performance is best with smaller responses Return LLM-friendly dataTools can

6、return unstructured or structured data.Combine common sets of actions into one callAvoid making LLMs repeatedly produce the correct orchestration of common sequences of service calls9MCP ToolsLibertyMCP Host an AI app that uses MCP ClientsMCP Client a client to connect to an MCP ServerMCP Server a s

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据《Data》标记内容,全文主要介绍了IBM TechXchange 2025会议中关于Liberty Model Context Protocol (MCP)的讨论。以下是关键点: 1. **MCP概述**:MCP是一个开源标准,用于AI应用与支持工具之间的交互,允许AI应用访问工具、资源和提示。 2. **MCP服务器功能**:支持多种传输方式,包括Stdio和Streamable HTTP,并可作为工具注册表和工具调用者。 3. **MCP工具最佳实践**:建议工具描述清晰,参数明确,对输入差异容忍,并返回所需数据。 4. **MCP工具类型**:包括Liberty Host、Client和Server,以及工具T1和T2。 5. **MCP服务器版本**:mcpServer-1.0 beta版本包含在Open Liberty 25.0.0.11-beta中。 6. **MCP功能**:包括工具、资源、提示、完成和日志。 7. **Java API/SPIs**:提供声明工具、提供响应元数据、返回内容、提供结构化数据和支持取消操作等功能。 8. **安全性**:MCP使用OAuth 2.1进行授权,Liberty正在研究支持所有Liberty支持的认证和授权方法。
"MCP如何简化AI服务调用?" "Liberty MCP Server beta版亮点?" "AI应用如何高效使用MCP工具?"
客服
商务合作
小程序
服务号
折叠