MCPは存続しているが課題に直面

4 Min ReadMeghana Somasundara, UberEsther ShittuNEW YORK CITY -- Eighteen months after Anthropic introduced its Model Context Protocol and four months after the vendor donated it to the Agentic AI Foundation, which is part of the Linux Foundation, the AI community is grappling with challenges in using the open standard.When the generative AI vendor launched MCP as an open source standard to connect AI models and external data sources or tools, questions arose in the AI world whether the protocol could really stand as the standard. Despite skepticism, adoption of the tool grew with all major cloud providers accepting MCP as the standard for connecting AI agents. However, recently, there has been talk that MCP may be losing ground, and some AI vendors, including AI search provider Perplexity, have moved away from it. Problems such as context bloat, in which the model accepts excessive context, and the lack of synchronization among different MCP servers within an organization have led some to abandon the standard.Related:OpenAI Updates Agents SDK, Aims at Building Secure AgentsMCP as a FoundationHowever, other organizations are doing the opposite, trying to use the standard to scale AI within their organizations."MCP is a fundamental building line for accelerating the adoption of AI across many of these cases, amongst many tools, humans and machines," said Keith Basil, VP and GM at multinational open source software vendor SUSE, in an interview on Thursday at the MCP Developer Summit presented by the Agentic AI Foundation. He noted that the technology is still in its initial stages. "They just need to know how the strength is, that there is security there, the policy oversight, for what's happening is there," Basil continued, referring to organizations that use MCP. Building MCP servers requires a time commitment, Bharat Kotian, senior technical support manager at Protegrity, an enterprise data security platform vendor, said in an interview.While Protegrity is just starting to experiment with MCP, Kotian said the open standard is "bringing a lot of things together," helping teams to be more productive and saving time."There's a time investment," he added. "[Teams] have to adapt to it." Moreover, teams should ensure they maintain the security of sensitive data, he said.Uber’s ExperienceOne such organization that worked to adapt to MCP is Uber. During a keynote presentation at the conference, Uber engineers revealed that the organization used MCPs to scale its engineering services and AI agents."MCPs are not just important, they really are what make AI useful at Uber," Meghana Somasundara, agentic AI lead at Uber, said during a keynote. However, before the ride-hailing giant could scale its use of MCP, it needed to tackle some of the challenges its engineers were facing using the protocol, such as not having central guidance on how to deploy the MCP servers that developers and engineers were building.Related:As AI Infosec Woes Heighten, IBM Intros Autonomous Security ServiceOne way Uber tackled the problem was to build an MCP Gateway and Registry as a centralized control plane for all its MCP interactions, enabling consistency across the various MCP servers Uber uses.Other Hurdles, and Competitors While MCP appears to be part of the foundation of agentic AI, it is not the only AI standard on the market. For example, it competes with other standards, such as Google's Agent2Agent protocol. Some MCP users also share an understanding that the standard even rivals Anthropic's own Claude Skills -- folders of instructions, scripts and resources that the Claude foundation model uses to improve a specific task.However, according to one MCP maintainer, Clare Liguori, senior principal software engineer at AWS, MCP and Skills complement each other."We see a lot of folks who are describing in a skill how to use an MCP, if their tools aren't very self-describing, or if there are particular workflows of combining multiple tools," Liguori said in an interview.Related:Anthropic Tool Speeds up AI Agent Development for EnterprisesShe added that what she really sees enterprises struggling with is control."Things like MCP Registry, things like MCP Gateways, are super important for giving them those control points," she continued.Despite MCP users and maintainers working to address the challenges many are facing with the open standard, one critical aspect that needs to be tackled is ethics."Not a single talk has mentioned responsible, ethical AI at a governance level or for every tool, how they are approaching this area," said Rupa Dachere, CEO and president of Thrive-WiSE, an organization dedicated to helping women engineers, in an interview. "Are they really thinking about it?"For Mazin Gilbert, executive director of the Agentic AI Foundation, while safety and risks can be defined using protocols, each organization still has its own responsibility."Every company will be building their enterprise-ready final production system and deploying that," Gilbert said. “[Organizations] take the responsibility for ethical AI."About the AuthorNews Writer, AI BusinessEsther Shittu brings four years of expertise covering artificial intelligence technologies and industry trends. As co-host of the "Targeting AI" podcast, she talks to thought leaders and practitioners exploring critical AI developments. Previous to AI Business, she wrote for several publications including the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters.