kongjie 博阳 腾讯科技《AI未来指北》特约作者 萌萌 编辑 以下文章来源于腾讯科技 ，作者kongjie 博阳 腾讯科技 . 腾讯新闻旗下腾讯科技官方账号，在这里读懂科技！ 当下AI领域最炙手可热的概念，莫过于MCP。 MCP 指的是Model Context Protocol（模型上下文协议）。令人意外的是，一个协议系统的热度，甚至 盖过了OpenAI发布的最新模型，成为行业讨论的焦点。 MCP的本质：统一的工具调用协议 什么是MCP？ MCP是一种开放的技术协议，旨在标准化大型语言模型（LLM）与外部工具和服务的交互方式。你可以 把MCP理解成像是一个AI世界的通用翻译官，让AI模型能够与各种各样的外部工具"对话"。 为什么需要MCP？ 在MCP出现之前， AI工具调用面临两大痛点： 第一是接口碎片化：每个LLM使用不同的指令格式，每个工具API也有独特的数据结构，开发者需要为 每个组合编写定制化连接代码；第二是开发低效：这种"一对一翻译"模式成本高昂且难以扩展，就像为 每个外国客户雇佣专属翻译。 而MCP则采用了一种通用语言格式（JSON - RPC），一次学习就能与所有支持这种协议的工具进行交 ...

AI"新晋顶流"！ 4月29日，阿里发布并开源了新一代通义千问模型Qwen3，其不仅以更高的性能、更低的成本而备受关注，而且 还强化了对MCP（Model Context Protocol，模型上下文协议）的支持。这意味着开发者可以基于Qwen3大模型 轻松调用各类外部数据源与工具，实现更低成本、高效地开发Agent（智能体）。 一些业内人士已经迫不及待地用Qwen3构建应用了。火星电波创始人、前MiniMax海螺AI产品负责人冯雷创建 了一个网页，成功地调用了图片、音频、地图等各个MCP，很快就实现了提示词的效果。他在其个人社交平台 上也分享了这一成果，并表示"MCP的原生支持非常重要"。 无独有偶，在4月25日举行的Create 2025百度AI开发者大会上，百度创始人李彦宏表示："MCP让AI更懂外部世 界、更容易获得信息、更自由地调用工具。我们认为MCP是AI发展的一大步，开发者要尽早地了解它、拥抱 它。"大会期间，百度智能云正式发布国内首个企业级MCP服务，支持开发者全面拥抱MCP。 简单理解，MCP可以视其为AI 时代的"万能插座"。通过MCP，大模型可以轻松调用各类外部数据源和工具，实 现与外界 ...

Core Insights - The emergence of the Model Context Protocol (MCP) is seen as a significant advancement in AI technology, enabling AI to connect with external tools and data sources for task execution rather than just conversation [1][3] - Baidu is the first to implement MCP in the e-commerce sector, aiming to reshape the business model by providing an "all-in-one" interface for AI applications [1][5] Industry Impact - MCP, proposed by AI company Anthropic, serves as an open standard that connects large models with external data sources, lowering development barriers in AI application [3][4] - The focus in the AI industry is shifting from model parameters to practical application deployment, with MCP facilitating this transition [3][4] Baidu's Innovations - Baidu's e-commerce platform, Baidu Youxuan, has integrated MCP capabilities, allowing AI agents to perform comprehensive tasks from product recommendations to completing purchases [7][11] - The platform utilizes a self-developed e-commerce knowledge graph to provide AI with structured information about products, enhancing decision-making accuracy [7][8] - Baidu Youxuan's MCP supports a full transaction cycle, enabling AI to automate the shopping process, which is a first in the domestic e-commerce sector [8][11] Developer Ecosystem - Baidu has established a robust support system for developers, providing technical documentation and tools to facilitate the integration of MCP into their applications [9][11] - Developers can leverage Baidu's extensive traffic ecosystem to promote their MCP-enabled applications, enhancing user acquisition [9][11] Strategic Positioning - Baidu's decision to launch MCP services in e-commerce reflects a strategic insight into changing user behaviors and the evolving flow of online traffic [12][14] - The integration of MCP with Baidu's search capabilities aims to reclaim its position as a primary entry point for users seeking products, thereby enhancing user engagement and retention [15][16] - The MCP framework offers a new monetization pathway for AI capabilities, allowing Baidu to earn commissions from transactions facilitated by AI [16][17]

Core Viewpoint - The article highlights the rapid rise of the Model Context Protocol (MCP) in the AI industry, particularly following Baidu's recent announcements and support for MCP, which is seen as a pivotal development for AI applications and developer engagement [1][3][20]. Summary by Sections MCP Overview - MCP stands for Model Context Protocol, an open standard introduced by Anthropic to unify communication between large language models and external data sources or tools [4][6]. - It serves as a "universal socket" for AI applications, allowing for standardized interfaces that facilitate interaction with various external tools and data sources [6][8]. Importance of MCP - Initially launched in November 2024, MCP did not gain much attention until early 2025 when several prominent applications began to adopt it, marking its transition to a de facto standard in AI interactions [9][10]. - The adoption of MCP has been driven by both developer acceptance and the integration of major model providers, including OpenAI, Google, Baidu, and others, which has significantly increased its visibility and utility [10][12]. Baidu's Strategic Moves - Baidu has committed to fully supporting MCP, launching the world's first e-commerce transaction MCP and search MCP services, thereby enhancing its AI ecosystem [12][19]. - The company aims to create a comprehensive MCP ecosystem, integrating its core platforms and applications to attract developers and partners [18][20]. Developer Benefits - MCP simplifies the development process for AI applications by reducing the need for multiple interface adaptations, thus increasing efficiency and lowering costs [14][15]. - Developers can now integrate various functionalities into their applications with minimal configuration, akin to building with blocks [14][15]. Future Outlook - The article suggests that the adoption of MCP could lead to a significant explosion in AI applications by 2025, similar to the mobile internet boom driven by technological advancements [15][20]. - Baidu's strategy positions it to compete effectively in the global AI landscape, leveraging its full-stack capabilities across various layers of AI technology [19][21].

Core Viewpoint - The release of the Model Context Protocol (MCP) service by Liou Digital marks a significant advancement for the "AI Pitcher" series, enhancing its capabilities in advertising campaign management and data analysis, thus evolving the "dual pitcher paradigm" into a new phase [1][3][22]. Group 1: MCP Service Introduction - The MCP service is an open standard introduced by Anthropic in 2024, designed to standardize interactions between AI models and external data sources or tools [3]. - The MCP service acts as a "universal toolbox" for the Liou AI Pitcher, enabling it to autonomously navigate various digital environments for advertising, data monitoring, and material management without human intervention [3][18]. Group 2: Upgrade of the Dual Pitcher Paradigm - Prior to the MCP service, the Liou AI Pitcher functioned primarily as a "helper" to human pitchers, lacking deep collaborative capabilities [5]. - With the introduction of the MCP service, the AI Pitcher can now perform complex tasks such as data collection, analysis, and report generation independently, significantly enhancing operational efficiency [5][15]. Group 3: Core Components of MCP - The MCP service comprises three core systems: identity authentication and security, programmatic advertising operations and data analysis interfaces, and internal system integration interfaces [18]. - The identity authentication system ensures secure access for the AI Pitcher, allowing it to operate within defined permissions and measure its output effectively [19]. - The programmatic advertising operations interface provides a comprehensive set of tools for managing advertising campaigns, budget control, and performance analysis, enabling real-time monitoring and automated optimization [20]. - The internal integration interface allows the AI Pitcher to interact with various internal systems, streamlining processes such as budget requests without manual input from human pitchers [21]. Group 4: Future Implications - The evolution of the Liou AI Pitcher and the introduction of the MCP service signify a rethinking of the human-AI relationship in marketing, paving the way for more AI employees to participate in advertising tasks [22]. - The successful integration of AI Pitchers into the marketing ecosystem represents a step towards achieving a truly collaborative "human-machine symbiosis" in the industry [22].

中金研究 MCP（Model context protocol）是由Anthropic于2024年11月提出的一套标准化交互协议。 2025年MCP讨论度攀升，核心源于OpenAI、Google等海外 头部厂商支持MCP协议，作为MCP Client方入局，当前阿里、腾讯、百度均积极拥抱MCP生态。 本文对MCP定义、意义、格局推演、未来路线及对软 件行业的影响进行展开。 点击小程序查看报告原文 Abstract MCP旨在为大模型及客户端提供标准化接口，使其高效、安全地调用外部数据源、工具，从而突破大模型的静态能力限制，为Agent提供技术底座和生态 支持。 MCP具备MIT开源许可，其开放属性和垂类场景适用性可类比为AI时代的"HTTP协议"或"USB-C"。类似LSP创建编辑器前端与语言后端解耦的标 准化通信架构，MCP核心价值在于解决LLM、Agent外部工具调用和集成的工作流和标准化问题，将M*N的难题简化成All-in-One。 MCP Server方兴未艾，数据库和搜索类厂商率先拥抱，垂类行业应用快速跟进，先拥抱者有望先受益于生态早期红利；MCP聚合平台迎早期流量红利。 MCP Server标准 ...

中国将 "适度" 减少美国电影进口数量。 国家电影局发言人称，美国政府加征关税 "会令国内观众对美国影片好感度进一步降低"，因此将 遵循市场规律，适度减少美国影片进口数量。中国是全球第二大电影市场，据猫眼娱乐数据，引入 国内的美国电影数量在 2018 年达到峰值，共 63 部，随后数量下滑，2022 年仅 25 部，去年又升至 42 部。近年来国产电影更受消费者支持，一个例子是《哪吒之魔童闹海》全球票房已破 156 亿， 进入全球影史票房第五。 优衣库大部分销往美国的优衣库产品都产自东南亚，此前特朗普宣布将对中国以外的其他国家、地 区暂停征收 90 天的关税。有分析师称，纺织供应链比其他行业供应链灵活，关税对优衣库的负面 影响将小于任天堂、丰田等日本公司。 马云现身阿里云谷园区。 社媒流传的影像显示，马云身穿阿里云 15 周年纪念 T 恤，并佩戴了阿里新版工牌，工牌上印有他 的照片和花名 "风清扬"。此行他参加了阿里云新财年启动会并上台讲话。据我们了解，阿里近期 在大力鼓励员工佩戴工牌，并推出了有戴脖、斜挎和卡夹三种戴法的新工牌，员工可自愿申请换 新，员工还可凭工牌享受园区附近商户的折扣。自去年底以来，马云已 ...

图片来源： Andreessen Horowitz Z Highlights: Yoko Li 是风投公司 Andreessen Horowitz （ a16z ）的合伙人，专注于开发者工具、基础设施与人工智能领域，是推动 AI 原生工具生态的重要推动者之一。 自从OpenAI于2023年推出Function Calling功能以来，我一直在思考，如何才能真正激活一个agent与工具协同运作的生态系统。随着基础模型变得越来越智 能，Agent与外部工具、数据和API的交互却日益碎片化：开发者需要为每一个系统单独编写特定的业务逻辑，才能让agent正常运行和集成。 很明显，我们迫切需要一个统一的标准接口（interface），来规范执行、数据获取和工具调用的方式。 API曾是互联网时代的重要桥梁，为软件之间的通信 建立了通用语言，而AI模型领域目前还缺乏类似的标准。 2024年11月推出的MCP（Model Context Protocol）在开发者（developer）和AI社区中快速引起关注，并有可能成为解决方案。 在这篇文章中，我们将介绍 什么是MCP，它如何改变AI与工具的交互方式，开发者已经基于它构建 ...