作者 | 陈仲寅 审校 | 李忠良 策划 | AICon 全球人工智能开发与应用大会 随着信息时代的到来，开发者和集成开发环境的概念才逐渐形成。当时，开发工具主要分为两类：文本编辑器和集成开发环境。文本编辑器如 Vim 和 Ultra Editor，并非专为编码设计。而以微软的 Visual Studio 为代表的集成开发环境则功能强大、集成度高。 此外，Java 领域出现了 Eclipse、NetBeans，JetBrains 的 WebStorm 和 IDEA 等工具，它们都以功能全面著称。 在演进的过程中，逐步延伸出了性能、安全、成本等当下的核心问题，而随着系统复杂度的提升和业务定制化需求增强，工具和 Agent 的生态能力逐渐 成为提升工程效率、增强平台能力的关键，是未来发展的基石。 本文整理自字节跳动 Trae 架构师陈仲寅在 AICon 2025 上海 的分享 " 打造可扩展的生态体系：从 MCP 到 Agent 集成的实践与趋势 "。本次分享将深入 探讨如何通过 MCP 以及 Agent 构建一个可扩展的生态体系，内容涵盖自定义 Agent 的设计与接入方式、如何与一方内部工具集成、MCP 等三 ...

专注于LLM+RL的科技公司OpenPipe提出 全新开源强化学习 框架——MCP·RL。 只需一个MCP Server的地址，agent就能自动发现工具、生成任务， 通过强化学习在闭环反馈中摸索出最优调用策略。 henry 发自 凹非寺 量子位 | 公众号 QbitAI 强化学习+任意一张牌，往往就是王炸。 在实测中，MCP·RL更是在 2/3的benchmark上达到或超过SOTA性能 ，效果直接拉满。 不套公式，在"做中学"，这就是专属RL的power！ MCP·RL的做中学 想明白MCP·RL怎么个"做中学"法，咱们有必要简单过一下传统MCP的流程： 举个例子，假如你想让agent帮自己读邮件、分类、写回复，那么你就得提前设置好整个工作流： 准备邮件数据、注册工具、写prompt规划执行顺序。 此外，你还得设置回退逻辑，以防中途崩掉。 而这只是一个发邮件的例子，功能一多，配置量指数级上升。 最关键的是——你得 知道怎么拆任务、调工具、写逻辑。 换句话说，agent就是在做你给他出的完形填空。 而你，我的朋友，要填除了空以外的所有东西。 MCP·RL的提出就是为了解决这一问题。 你只需提供MCP Ser ...

Core Viewpoints - The article discusses the concept of "intelligent agents" and their potential to transform AI applications, emphasizing the need for agents that can effectively assist users in completing tasks [2][3][13]. Group 1: Definition and Characteristics of Intelligent Agents - Intelligent agents are defined as systems that can assist or replace humans in completing specific tasks, characterized by capabilities such as memory, planning, execution, and reflection [5][9]. - The evolution of intelligent agents is driven by advancements in large models and the integration of various technologies, including RPA and API [6][14]. - The distinction between intelligent agents and traditional automation tools lies in their ability to autonomously plan and execute tasks rather than merely following predefined workflows [10][15]. Group 2: Market Trends and Product Forms - The article identifies two main forms of intelligent agents: those embedded within foundational large models and standalone agents that operate independently [18][19]. - The future of intelligent agents is expected to be shaped by their ability to connect with the physical world, making them essential for practical applications [14][17]. - The competition among different intelligent agents will likely focus on service quality, response speed, and pricing, marking a shift from traditional user interface-driven applications [17][19]. Group 3: Challenges in Implementation - The article highlights several challenges in the deployment of intelligent agents, including the need for clear task definitions and the ability to handle complex workflows [28][30]. - A significant portion of tasks in B2B environments is standardized, making them suitable for automation by intelligent agents, while more creative tasks remain challenging [29][30]. - The limitations of current intelligent agents in managing context and memory during task execution are noted as critical obstacles to their effectiveness [34][35]. Group 4: Future Outlook and Opportunities - The potential for intelligent agents to evolve into more versatile systems that can collaborate with other agents is discussed, suggesting a future where agents can autonomously find and utilize other agents to complete tasks [15][26]. - The article posits that while foundational large models may dominate certain applications, specialized agents will still be necessary for complex, industry-specific tasks [37][38]. - The ongoing development of intelligent agents is expected to create new opportunities across various sectors, particularly in automating routine tasks and enhancing productivity [39][40].

Core Viewpoint - The release of Kimi K2 marks a significant turning point for the company, indicating a shift from a reliance on scaling laws to a more innovative approach in AI model development and strategy [2][4][22]. Group 1: Kimi K2 Release and Its Impact - Kimi K2 achieved a global fifth ranking in the LMArena leaderboard and first among open-source models, surpassing competitors like Claude 4 and DeepSeek-R1-0528 [2]. - The release is seen as more than just a temporary success; it represents a deeper strategic shift for the company and the industry [4][22]. - Kimi K2 introduces two major advancements: an expansion of model parameters to over 1 trillion and the concept of "model as agent," allowing for tool utilization [23][35]. Group 2: Challenges Faced by Kimi - Kimi's previous strategy relied heavily on scaling laws, believing that larger models and more data would lead to better performance, but this approach faced challenges as high-quality data became scarce [8][13][14]. - The company's user growth strategy was questioned after competitors like DeepSeek demonstrated significant user acquisition without marketing spend, highlighting the need for a more effective product [18][54]. - Kimi's marketing budget reached approximately 900 million RMB in 2024, yet user engagement declined, indicating a disconnect between spending and user retention [17]. Group 3: Strategic Transformation - The company has shifted its focus from aggressive marketing to enhancing model performance and embracing open-source collaboration, reflecting a significant cultural change [55]. - Kimi's team has decided to halt all marketing activities and concentrate resources on foundational algorithms and the K2 model, emphasizing the importance of product quality over quantity [55]. - The strategic pivot is seen as a response to the success of DeepSeek, which has prompted Kimi to adopt more effective architectural choices and prioritize technical research [55][56].

AI Engineering Resources - A free illustrated guidebook on MCP fundamentals is available [1] - The guidebook contains over 75 pages [1] - The guidebook includes 11 hands-on projects for AI engineers with code [1] MCP Fundamentals - The guidebook visually explains MCP fundamentals [1] - The approach is 100% hands-on [1]

Open Source LLM Framework - A framework connects any LLM to any MCP server (open-source) [1] - The framework enables building custom MCP Agents without closed-source apps [1] - Compatible with Ollama, LangChain, etc [1] - Allows building 100% local MCP clients [1]

Agentic Economy & MCP Standard - The agentic economy is emerging, where AI agents can interact, find counterparts, and purchase services from other agents, businesses, or tools [4] - MCP (Message Communication Protocol) is becoming a standard for agentic interaction, dominating the space compared to Open API and Google's A2A [8][9] - Tool discovery, a key feature of MCP, allows agents to dynamically discover and use tools based on the workflow, differentiating it from Open API [7][8] - A centralized marketplace of MCP services, like APIFY, can provide access to various services with a single API token, enabling rapid scaling of the ecosystem [12] APIFY's Role & Marketplace - APIFY is a marketplace of 5,000 tools (actors), primarily data extraction tools, with a community of creators who monetize their tools [4] - Actors are self-contained software units with defined input and output, facilitating easy integration with other systems [4][5] - APIFY has integrations with workflow automation tools and MCP, enabling AI agents to call actors from the marketplace [6][7] - APIFY enables publishing and monetization of tools or agents, providing access to a broad ecosystem of developers and visibility [23][24] Challenges & Future - Agents currently rely on human developers for access to tools and services, hindering their ability to autonomously find and purchase services [10][11] - Trust between agents and tools is a key open question, as is the overall value and reliability of autonomous tool discovery [25][26][27] - The company paid out over $4 million to creators last month, with actors generating over $500,000 per month, indicating rapid ecosystem growth [23]

MCP Overview & Architecture - MCP (Micro Control Plane) is defined as a pluggable architecture for agents, contextualized within an enterprise cloud service [5][6] - Sentry's MCP server was initially built as a fun project and is biased towards Sentry's application monitoring services [4][5] - The industry views MCP as a potential solution for integrating services into various agents, enabling bug fixes and workflow enhancements within editors [7][8][25] Implementation & Challenges - Implementing MCP involves complexities around OAUTH 21%, requiring solutions like Cloudflare Shim for proxying OAUTH 2 API [16][17] - A key challenge is that MCP cannot simply sit on top of Open API; systems need to be designed around how agents and models react to provided context [19][20][21] - Current client support for native authentication is still evolving, with some clients like Cursor experiencing breakage [22] Security & Best Practices - Security is a major concern, particularly with the standard IO interface, and random MCP tools should not be allowed within organizations [27] - For B2B SaaS companies, focusing on OAUTH with remote environments is crucial for integrating services into agents [25] - Companies should avoid simply proxying Open API and exposing it as tools, as this yields poor results; intentional design and context provision are necessary [30] Agent-Centric Approach - The industry should focus on building agents, viewing MCP as a plug-in architecture to leverage the value of LLMs [39][40] - Exposing agents through the MCP architecture, particularly in B2B settings, is seen as a significant value unlock [42] - Optimizing for context in workflows and understanding data is crucial when designing agents, with a focus on providing structured information like Markdown for language models [31][50]

Core Viewpoint - The article discusses the significant changes in Baidu's search engine, marking it as the largest revision in a decade, emphasizing the integration of AI technologies into search functionalities and the potential implications for investment and AI entrepreneurship [2][3]. Group 1: Changes in Search Engine - Baidu has established itself as synonymous with "search" in the Chinese market, achieving daily search volumes in the billions [2]. - The traditional search engine business model heavily relied on online marketing services, particularly advertising revenue, which accounted for over half of its income [3][4]. - The advent of AI transforms search capabilities, allowing for more natural language processing and user intent understanding, moving beyond simple keyword matching [5][8]. Group 2: New Features and Capabilities - The updated search interface allows for longer, more conversational queries, accommodating a wider range of user inputs [10]. - Multi-modal input methods have been introduced, including voice search and image recognition, enhancing user interaction [15][19]. - Baidu's AI search now generates results that include text, images, and videos, with a focus on providing credible sources for the information presented [23][26]. Group 3: Ecosystem and Collaboration - The introduction of the MCP (Model Capability Protocol) facilitates collaboration between various AI applications, positioning Baidu as a leader in integrating AI capabilities across platforms [26]. - Baidu's search platform has integrated with 18,000 MCPs and over 220 AI applications, creating a diverse and open ecosystem [26]. Group 4: Video Generation and AI Creativity - Baidu has launched the MuseSteamer video generation model, which has gained recognition for its performance in generating high-quality videos from images [31]. - The model supports the creation of videos with sound, enhancing the creative possibilities for content creators [31]. Group 5: Future Implications - The changes in Baidu's search engine represent a significant shift towards a more integrated AI ecosystem, with the potential to redefine user experience and engagement [33]. - The competition among major tech companies to dominate the AI landscape is intensifying, presenting both opportunities and challenges for investors and entrepreneurs [32][33].

RT Avi Chawla (@_avichawla)After MCP, A2A, & AG-UI, there's another Agent protocol (open-source).ACP (Agent Communication Protocol) is a standardized, RESTful interface for Agents to discover and coordinate with other Agents, regardless of their framework (CrewAI, LangChain, etc.).Here's how it works:- Build your Agents and host them on ACP servers.- The ACP server will receive requests from the ACP Client and forward them to the Agent.- ACP Client itself can be an Agent to intelligently route requests to t ...