How Cursor’s New AI Features Revolutionize Code Review and Remote Development

Explosive news: after two and a half years, Cursor finally releases version 1.0.

As an avid user, I tested the major updates: BugBot for code review, the first memory feature, one‑click MCP setup, Jupyter support, and the fully opened Background Agent.

1. Using BugBot for Automatic Code Review

In team development, each code change is submitted as a pull request (PR) that needs review before merging.

BugBot leverages AI to automatically review PRs and catch potential errors.

Enable BugBot in the web version settings, authorize the desired GitHub repositories.

Note that AI model calls are billed per token; roughly 1 CNY per 1,000 characters.

Set a monthly spending limit and configure options such as disabling automatic execution.

Testing with a buggy Java file shows BugBot leaving a comment with the detected issues.

Click “Fix in Cursor” to jump to the editor with a pre‑filled fix suggestion.

2. Background Agent Open to All Users

Background Agent, previously in beta, is now officially available to everyone, providing a cloud server that runs AI tasks remotely.

Disable privacy mode in settings to use it.

Open the Agent panel with Cmd/Ctrl+E, then enter a task. The agent requires an initialized Git repository.

Install Git, run git init in Cursor’s terminal, and link a remote GitHub repository.

After linking, restart Cursor, then the Agent can execute tasks even after the local machine is shut down.

The Agent panel shows the AI working on a cloud server, saving local resources.

Future possibilities include shared agents for team collaboration.

Background Agent still uses the Max model and is billed per token; compute resources may become a paid feature.

3. Memory Feature (Memories)

The new memory feature lets Cursor remember specific rules from user‑AI conversations and reuse them later.

It differs from chat history; only selected rules are stored, e.g., “prefer Windows commands when generating code.”

Rules can be managed or deleted in the settings page.

4. One‑Click MCP Installation and OAuth Support

MCP (Model‑Component‑Package) provides a standard way to add AI‑enhanced tools.

Previously, adding an MCP service required editing JSON configuration files manually. Now, the official tool list lets you install MCP services with one click and adjust authentication.

Some services also support OAuth for quick web‑based authentication.

Although the current selection is limited, the workflow is much smoother.

Other Improvements

Additional updates include richer chat responses with visualizations (Mermaid diagrams, Markdown tables), Jupyter Notebook Agent support, a revamped settings dashboard with usage analytics, and various UI and performance tweaks.

Keyboard shortcut: Cmd/Ctrl+E opens the Background Agent panel. @Link and web search can now parse PDFs and include them in context.

Network diagnostics in settings to verify connectivity.

Parallel tool calls for faster responses.

Collapsible tool calls in chat.

Account updates: enterprise users access only stable releases; team admins can disable privacy mode; new management API for usage metrics.

Model update: Max mode is now available for Gemini 2.5 Flash.