March 2025 (version 1.99)
Join VS Code at Microsoft Ignite or catch the sessions online. A year of VS Code, wrapped. Explore the highlights. March 2025 (version 1.99) Update 1.99.1 : The update addresses these security issues . Update 1.99.2 : The update addresses these issues . Update 1.99.3 : The update addresses these issues . Downloads for 1.99.3 Windows x64 Arm64 macOS Universal Intel Apple silicon Linux .deb .rpm .tar.gz Arm instructions Snap Welcome to the March 2025 release of Visual Studio Code.
There are many updates in this version that we hope you'll like, some of the key highlights include: Agent mode Agent mode is available in VS Code Stable. Enable it by setting chat.agent.enabled Open in VS Code Open in VS Code Insiders ( more... ). Extend agent mode with Model Context Protocol (MCP) server tools ( more... ). Try the new built-in tools in agent mode for fetching web content, finding symbol references, and deep thinking ( more... ).
Code editing Next Edit Suggestions is now generally available ( more... ). Benefit from fewer distractions such as diagnostics events while AI edits are applied in the editor ( more... ). Chat Use your own API keys to access more language models in chat (preview) ( more... ). Easily switch between ask, edit, and agent mode from the unified chat experience ( more... ). Experience improved workspace search speed and accuracy with instant remote workspace indexing ( more... ).
Notebook editing Create and edit notebooks as easily as code files with support for edit and agent mode ( more... ). If you'd like to read these release notes online, go to Updates on code.visualstudio.com . Insiders: Want to try new features as soon as possible? You can download the nightly Insiders build and try the latest updates as soon as they are available. Chat Agent mode is available in VS Code Stable Setting : chat.agent.enabled Open in VS Code Open in VS Code Insiders We're happy to announce that agent mode is available in VS Code Stable!
Enable it by setting chat.agent.enabled Open in VS Code Open in VS Code Insiders . If you do not see the setting, make sure to reload VS Code. Enabling the setting will no longer be needed in the following weeks, as we roll out enablement by default to all users. Check out the agent mode documentation or select agent mode from the chat mode picker in the Chat view. Model Context Protocol server support This release supports Model Context Protocol (MCP) servers in agent mode.
MCP offers a standardized method for AI models to discover and interact with external tools, applications, and data sources. When you input a chat prompt using agent mode in VS Code, the model can invoke various tools to perform tasks such as file operations, accessing databases, or retrieving web data. This integration enables more dynamic and context-aware coding assistance. MCP servers can be configured under the mcp section in your user, remote, or .code-workspace settings, or in .vscode/mcp.json in your workspace.
The configuration supports input variables to avoid hard-coding secrets and constants. For example, you can use ${env:API_KEY} to reference an environment variable or ${input:ENDPOINT} to prompt for a value when the server is started. You can use the MCP: Add Server command to quickly set up an MCP server from a command line invocation, or use an AI-assisted setup from an MCP server published to Docker, npm, or PyPI. When a new MCP server is added, a refresh action is shown in the Chat view, which can be used to start the server and discover the tools.
Afterwards, servers are started on-demand to save resources. Theme: Codesong (preview on vscode.dev ) If you've already been using MCP servers in other applications such as Claude Desktop, VS Code will discover them and offer to run them for you. This behavior can be toggled with the setting chat.mcp.discovery.enabled Open in VS Code Open in VS Code Insiders . You can see the list of MCP servers and their current status using the MCP: List Servers command, and pick the tools available for use in chat by using the Select Tools button in agent mode.
You can read more about how to install and use MCP servers in our documentation . Agent mode tools This milestone, we have added several new built-in tools to agent mode. Thinking tool (Experimental) Setting : github.copilot.chat.agent.thinkingTool Open in VS Code Open in VS Code Insiders . Inspired by Anthropic's research , we've added support for a thinking tool in agent mode that can be used to give any model the opportunity to think between tool calls.
This improves our agent's performance on complex tasks in-product and on the SWE-bench eval. Fetch tool Use the #fetch tool for including content from a publicly accessible webpage in your prompt. For instance, if you wanted to include the latest documentation on a topic like MCP , you can ask to fetch the full documentation (which is conveniently ready for an LLM to consume) and use that in a prompt. Here's a video of what that might look like: In agent mode, this tool is picked up automatically but you can also reference it explicitly in the other modes via #fetch , along with the URL you are looking to fetch.
This tool works by rendering the webpage in a headless browser window in which the data of that page is cached locally, so you can freely ask the model to fetch the contents over and over again without the overhead of re-rendering. Let us know how you use the #fetch tool, and what features you'd like to see from it! Fetch tool limitations: Currently, JavaScript is disabled in this browser window. The tool will not be able to acquire much context if the website depends entirely on JavaScript to render content.
This is a limitation we are considering changing and likely will change to allow JavaScript. Due to the headless nature, we are unable to fetch pages that are behind authentication, as this headless browser exists in a different browser context than the browser you use. Instead, consider using MCP to bring in an MCP server that is purpose-built for that target, or a generic browser MCP server such as the Playwright MCP server .
Usages tool The #usages tool is a combination of "Find All References", "Find Implementation", and "Go to Definition". This tool can help chat to learn more about a function, class, or interface. For instance, chat can use this tool to look for sample implementations of an interface or to find all places that need to be changed when making a refactoring. In agent mode this tool will be picked up automatically but you can also reference it explicitly via #usages Create a new workspace with agent mode (Experimental) Setting : github.copilot.chat.newWorkspaceCreation.enabled Open in VS Code Open in VS Code Insiders You can now scaffold a new VS Code workspace in agent mode .
Whether you’re setting up a VS Code extension, an MCP server, or other development environments, agent mode helps you to initialize, configure, and launch these projects with the necessary dependencies and settings. VS Code extension tools in agent mode Several months ago, we finalized our extension API for language model tools contributed by VS Code extensions. Now, you can use these tools in agent mode. Any tool contributed to this API which sets toolReferenceName and canBeReferencedInPrompt in its configuration is automatically available in agent mode.
By contributing a tool in an extension, it has access to the full VS Code extension APIs, and can be easily installed via the Extension Marketplace. Similar to tools from MCP servers, you can enable and disable these with the Select Tools button in agent mode. See our language model tools extension guide to build your own! Agent mode tool approvals As part of completing the tasks for a user prompt, agent mode can run tools and terminal commands.
This is powerful but potentially comes with risks. Therefore, you need to approve the use of tools and terminal commands in agent mode. To optimize this experience, you can now remember tha
The rest of this page is on the live site.