VS Code v1.103: AI Chat Checkpoints & Enhanced MCP Tools
The July 2025 update to Visual Studio Code, version 1.103, marks a significant stride in integrating artificial intelligence seamlessly into the developer workflow, introducing “Chat Checkpoints” and substantial enhancements to the “Model Context Protocol (MCP) tooling.” These features are set to empower developers with greater control, reliability, and configurability within their AI-driven coding environments.
A standout new addition is Chat Checkpoints, a feature designed to provide developers with a crucial safety net when working with AI-generated code. This functionality allows users to restore different states of their chat conversations, effectively enabling them to revert edits and rewind to specific points in their AI-driven coding sessions. This capability proves particularly invaluable in scenarios where an AI agent might propose changes across multiple files, offering a streamlined way to undo a series of modifications. By capturing snapshots of the workspace and chat history, VS Code ensures that developers can experiment with AI suggestions confidently, knowing they can easily revert to a previous, stable state without relying on manual Git commits for every iteration. This addresses a long-standing desire for granular control over AI-driven changes, echoing similar capabilities found in other AI-first code editors. The feature is enabled by default, though it can be managed via the chat.checkpoints.enabled setting.
Further bolstering VS Code’s AI capabilities are the improvements to the Model Context Protocol (MCP) tooling. MCP, an open standard that draws inspiration from the widely adopted Language Server Protocol (LSP), provides a standardized method for AI models to interact with external tools, applications, and data sources. This protocol is fundamental to VS Code’s “agent mode,” where AI functions as an autonomous pair programmer. In this mode, the AI can execute multi-step coding tasks, analyze codebases, suggest modifications, run terminal commands, monitor for errors, and iteratively resolve issues. The power of agent mode is significantly extended by its ability to leverage tools contributed by MCP servers, alongside built-in VS Code tools and extensions.
The v1.103 update specifically refines this MCP integration, introducing a revamped tool picker experience for more intuitive selection and management of available tools. Crucially, it also enables the use of more than 128 tools per agent request through an experimental tool-calling mode that intelligently groups tools when the limit is exceeded. These enhancements collectively boost the overall capacity, reliability, and configurability of AI agents operating within VS Code.
The maturation of MCP has been a progressive journey, with its support moving out of preview and becoming generally available in VS Code version 1.102 in June 2025. The latest update further solidifies this foundation by supporting the complete MCP specification, which now encompasses authorization, dynamic prompts, resources, and sampling. This evolution transforms MCP from a collection of individual tools into a comprehensive standard for integrating AI agents. A significant advancement is the new authorization specification, a collaborative effort involving Microsoft, Anthropic, and leading identity providers, which allows developers to delegate authentication to existing identity providers, ensuring enterprise-grade security for remote MCP servers. This robust framework enables AI assistants like GitHub Copilot to seamlessly access external resources such as file systems, databases, and APIs, making their suggestions more informed and contextually relevant.
This latest VS Code update underscores a clear commitment to empowering developers with sophisticated, yet manageable, AI-powered tools. By enhancing the ability to rewind AI-driven sessions and expanding the interoperability and reliability of AI agents through improved MCP tooling, Microsoft is refining the developer experience, making AI assistance more robust, controllable, and deeply integrated into the modern coding workflow.