VS Code 1.101 increases the security of MCP and improves version control
The new version of Visual Studio Code completes the MCP protocol and thus improves security and functions for developers.
(Image: Zakharchuk/Shutterstock.com)
With the update 1.101 (“May”) of Visual Studio Code (VS Code), the editor masters the complete Model Context Protocol (MCP) with all specifications: Authorization increases security when using MCP sources and sampling enables communication with the LLM.
The MCP protocol was only updated in March, introducing authorization between client and server using an external identity management system such as OAuth. This removes the responsibility from developers to build in their own secure authorization mechanisms for MCP sources. And they can drive the development of the server forward independently of user management. As an example, the blog entry mentions the GitHub MCP server, which can be used in VS Code with the GitHub credentials that are usually available there anyway. Users only need to grant access once, as the following image shows.
(Image: Microsoft)
Communication between server and LLM
Other new MCP features in VS Code are sampling, prompts and resources. With sampling, an MCP server can access the LLM directly and use the API key of the editor (as a client). The server does not need its key for the LLM. Prompts are prompt templates for the LLM, which the server generates dynamically and which users can pass on to the LLM according to the current context. Resources serve as additional sources, for example screenshots or log files.
Developers can now debug MCP servers, which currently only works with Python or Node.js. They also have the option of publishing multiple servers in one extension.
Videos by heise
Resources in the version control
The version control has a new view: In the list of previous commits, the changed files can now be expanded and displayed specifically in the editor on the right. VS Code shows the changes highlighted in color, as in the diff view. This helps with visual searches. Older commits can now be used for co-pilot chats, for example, to have changes explained.
(Image: Screenshot Microsoft)
Independently of version control, there is a new function for the AI chat: tool sets, in which users bundle several tools that the copilot processes step by step in a joint task.
(who)
