Claude introduces the "MHS Protocol," a robot version of the "MCP Protocol," to facilitate interaction between large models and robots.

Anthropic is replicating its success in software interoperability in the physical world. The AI startup has launched a new software specification called the "Model Hardware Standard" (MHS), designed to enable AI assistants like Claude to work more efficiently with robots, scientific instruments, and manufacturing hardware, marking a key step in the penetration of large models into the physical world.

According to Bloomberg, Anthropic announced on Thursday (August 27) the release of MHS as a "research preview," open to developers in the science, robotics, and manufacturing sectors for testing. This allows them to verify how Claude interacts with the physical world and establish necessary security mechanisms before official rollout. Currently, industry participants can join a waiting list to apply for trial use. Anthropic stated that it will open-source the framework after the preview period ends, but did not disclose a specific timeline.

The launch of MHS comes at a time when the commercial value of integrating AI with the physical world is gaining widespread attention. A previous report by Barclays predicted that the AI-driven robotics and autonomous machine market is expected to grow into a trillion-dollar sector by 2035. Google, OpenAI, and Nvidia are also simultaneously developing AI models and software for the robotics field.

MHS's core logic: Enabling AI to "understand" hardware.

The essence of MHS is to provide AI assistants with a standardized information storage mechanism for understanding how hardware works, thereby replacing paper manuals or tacit knowledge that only exists in the minds of a few experts.

Taking a factory robotic arm as an example, hardware manufacturers can explicitly define the safety boundaries for AI control of the robotic arm in the MHS specification, such as limiting the speed or angle of movement, thereby mitigating safety risks while granting AI control capabilities. Anthropic engineer Alek Kemeny stated:

"What MCP does for the software world, MHS will implement for the hardware world. It is especially valuable in scientific laboratory settings—where dozens or even hundreds of devices are often running simultaneously."

Benchmarking against MCP: From software interconnect to hardware interconnect

MHS's strategic logic is consistent with Anthropic's previously launched Model Context Protocol (MCP). MCP has become the industry standard for interaction between AI assistants and third-party software applications, enabling Claude and its competitor ChatGPT to connect to mainstream applications such as Gmail, Google Calendar, and Slack.

Anthropic launched MHS with the positioning of "hardware version of MCP", attempting to replicate the success of MCP in the field of physical device interconnection - by formulating open standards, building a hardware ecosystem moat around Claude, and promoting the standard to become an industry standard after open source.

In addition, Anthropic revealed that MHS has completed early testing with several industry organizations, including Amazon Web Services (AWS), Danaher, Hugging Face, and Raspberry Pi.

According to reports, in a promotional video released by Anthropic, a scientist from the biotechnology company Genentech sent Claude a PDF document of a self-designed experiment. Claude then independently called upon hardware equipped with MHS specifications to complete the entire experimental process.

Jonah Cool, Head of Collaboration and Deployment for Life Sciences at Anthropic, stated:

"Often, scientific research stalls because researchers don't know how to use the equipment, or the operation is too difficult. MHS enables Claude to become an empowering tool for scientists, helping them use the equipment correctly and professionally."

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.