Cherry Studio 2.1.0 adds website WebMCP support, expanding browser control and document section references

Cherry Studio released version 2.1.0 on September 18, enabling agents and chat assistants to control its built-in browser and adding support for website WebMCP tools. The release can locate page elements, recover element references after page updates, and inspect console and network activity, providing additional diagnostic clues when web interactions fail. [Release notes](https://github.com/CherryHQ/cherry-studio/releases/tag/v2.1.0)

WebMCP’s technical significance lies in allowing websites to expose existing JavaScript functionality as tools with descriptions and parameter schemas. Agents can then invoke explicit operations on supported websites, potentially reducing the need to locate interface elements step by step. However, websites must actively expose these tools. At the time of this review, the specification remained a community draft, not yet a W3C standard, so interoperability across arbitrary websites cannot be assumed. [WebMCP specification](https://webmachinelearning.github.io/webmcp/)

The specification also cautions that if a tool is unregistered and immediately registered again under the same name with a different schema, old arguments may be applied to the new tool. This raises synchronization issues between the page lifecycle and tool calls. Integrators should validate failure handling during navigation, reloads, and tool list updates, rather than testing only a single successful demonstration. [Race condition discussion in the specification](https://webmachinelearning.github.io/webmcp/)

The same release adds structured document selection: users can specify worksheet ranges, Word paragraphs, PDF pages, or slides, then pass them to the office-transform skill for extraction into new files. For workflows involving report preparation, this defines a clearer scope than “process this document.” Whether formulas, formatting, and cross-page relationships are preserved still needs to be validated with actual files. The minimum macOS requirement has also been raised to Ventura 13, making compatibility checks necessary before deployment on older machines. [Version changes](https://github.com/CherryHQ/cherry-studio/releases/tag/v2.1.0)

As more tools become available, permission settings require closer review for each execution mode. The official documentation lists modes such as confirmation for each action, automatic acceptance of edits, and smart approval, but Pi and DeepSeek Harness offer different options. External channels can also override agent settings. Passing desktop tests therefore does not establish that the same agent has identical operational boundaries when launched from a chat channel. [Permissions documentation](https://docs.cherryai.com.cn/advanced-basic/agent-workspace/permissions-memory-background)

The verifiable gains in this update are expanded interfaces for operations and document references; the official release has not yet provided task success rates across websites. Teams adopting the release should next record whether tasks can resume after page reloads, whether failed tool calls are retried, and how operations with side effects are reviewed. Regression tests using a fixed set of websites and document samples can then help determine whether the new capabilities reduce overall maintenance costs.