Getty Images/iStockphoto
Anthropic's new standard raises AI privacy, other concerns
The Model Context Protocol helps developers connect data to assistants. Some have concerns about users' data and whether a single vendor should create such a standard.
AI foundation model provider Anthropic proposed a new framework for connecting AI assistants to their data.
The generative AI vendor, developer of the Claude family of large language models and an OpenAI rival, introduced the Model Context Protocol on Nov. 25. Anthropic said MCP provides a universal open standard for connecting AI systems with data sources.
MCP enables developers to either expose their data through MCP servers or build AI applications that connect to the servers. Three major components of MCP are the Model Context Protocol specification and SDKs, local MCP server support in the Claude desktop apps, and an open source repository of MCP servers.
Anthropic said MCP is open source and aims to help frontier models produce better responses.
Anthropic's introduction of MCP comes as agentic AI continues to gain traction in the AI market, and more vendors are introducing autonomous and semiautonomous AI agent technology. With agentic AI and LLMs, developers need to streamline the process of connecting data to their AI agents or assistants.
More vendors needed
However, said Mark Beccue, an analyst at TechTarget's Enterprise Strategy Group, introducing a standard is not something a standalone vendor can do.
Mark BeccueAnalyst, Enterprise Strategy Group
"A standard is when the community agrees that this is what we're going to do, and it's collaborative," Beccue said. "They open sourced this, but they're the only ones talking about it. I just think it's a nonstarter."
He added that what Anthropic might be looking for with MCP is a standardized way to enable LLMs to talk to proprietary data. Anthropic might then need to start a conversation with other vendors about what standard is needed.
However, a standard would be better if it came from a group like the AI Alliance, which was started by IBM, Meta and other AI vendors, Beccue argued. This is because it would be the decision of a group of vendors rather than just one.
Anthropic said in a blog post that it is committed to making MCP a collaborative open source project and is looking to receive feedback.
" Unlike other AI labs, Anthropic doesn't train our generative AI models on user-submitted data by default," according to an Anthropic spokesperson.
Developers and the question of data
Nevertheless, MCP is helpful for developers looking for a streamlined process of connecting AI models to their data, said Sean Ren, an associate professor of computer science at the University of Southern California and CEO of Sahara AI, vendor of an AI blockchain platform.
"[Anthropic] allows you to follow the protocol to connect various resources into the large models and apply tools on top of them to achieve application," Ren said. "This is more like a framework."
While MCP could help developers, Anthropic did not address how it will protect user privacy and data security, Ren said. He noted that while MCP is open source, users still access it through Anthropic's closed model, Claude.
"Individual businesses need to be very careful about using the protocol, because you use the protocol to connect your personal data to the Anthropic models," he said. "Essentially, you are giving them access to your very private data, and there are no clear details about how they're going to protect those data, whether they're going to access it for other uses."
Anthropic also revealed on Nov. 26 that users of Claude.ai, the web version of the LLM, can now tailor responses to match preferences, tone and structure. Users can pick from these styles: formal, concise and explanatory.
Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.