Anysphere has released version 1.0 of its AI editor, Cursor, along with new features including previews of a BugBot that automatically reviews pull requests (PRs) and finds issues, a Background Agent that edits code in a remote environment, and a Memories feature which stores and references facts from previous AI chats.
BugBot connects to GitHub and inspects the content of code commits. It is triggered automatically when a PR is updated and leaves comments about issues that it finds, with a link to open the relevant code in Cursor. Use of BugBot requires Cursor’s Max mode, which has widened context and tools and requires at least a Pro subscription ($20 per month).
Background agents, currently in beta, allows developers to create agents which run in a remote environment provisioned by Cursor. The agent will clone a GitHub repository, work on a separate branch, and push changes. The docs state that the remote environments may become chargeable in future, but currently only AI token usage is charged. Like BugBot, background agents require Max mode.
The docs also warn that background agents have “a much bigger surface area of attacks compared to existing Cursor features,” and confesses that “our infra has not yet been audited by third parties.” The risks come about because agents have read-write privileges to repositories, and they both auto-run all commands, and run code on Cursor’s infrastructure. “This opens up the door for certain kinds of prompt injection attacks,” the docs state. For example, a malicious web page may be found by the agent that appears to help solve a problem but also contains instructions which exfiltrate code or secrets.
We recently highlighted the issue of prompt injection as a security issue that has no obvious fix, and which becomes more risky as agents gain more capabilities via MCP (model context protocol) servers.
MCP servers now have a one-click install in Cursor. The company is maintaining a list of approved MCP tools, including GitHub, Figma, Notion, Stripe and Playwright, and it is also possible for developers of MCP servers to add a deep link to Cursor in their documentation.
Another new feature, also in beta, is called Memories. During a chat, developers may give instructions or establish facts that can now be stored for future reference. Memories can be enabled, managed and viewed in settings.
Cursor is a fork of Visual Studio Code (VS Code) and was first previewed in 2023. It is one of a rapidly evolving group of tools from multiple vendors, with others including VS Code with GitHub Copilot, Windsurf (formerly Codeium), and AI plug-ins for VS Code and/or JetBrains IDEs such as Amazon Q Developer, Google Gemini, Mistral Code, Tabnine and more. Anthropic’s Claude Code runs in the terminal but also has integration with IDEs. Aider also runs on the command line but can respond to prompts in an IDE by watching files. Many of these products also work in browser-based environments.
The pace of change in this crowded market makes it challenging for developers to know which, if any, AI assistants they should adopt. Microsoft’s GitHub was early to market with Copilot, and has the advantage of first-party integration with VS Code. Claude Code is winning praise for the quality of its AI assistance. Developers in this discussion suggest that Cursor is at risk of being left behind by Claude Code, though some are happy with it. Forking VS Code is also an issue; companies like Anysphere have good reason not to use Microsoft’s official distribution; but the popularity of VS Code and the extent of its extension ecosystem makes these forks less attractive. “I’m only going to consider AI Tools that integrate with my IDEs,” said a comment on Cursor 1.0.