CLI Agent Modes and Cloud Handoff

Agent Modes in CLI

Plan Mode

Cursor brought Plan Mode to the CLI, allowing developers to sketch out their coding strategy before writing any code. Accessible via the /plan slash command or the --mode=plan flag, Plan Mode enables developers to work through their approach systematically.

In Plan Mode, the agent asks clarifying questions to refine the strategy and ensure all requirements are understood before implementation begins. This planning phase helps developers think through edge cases, identify potential challenges, and establish clear implementation paths.

Ask Mode

Ask Mode arrived in the CLI to support exploratory workflows where developers need to understand codebases without making modifications. Triggered using /ask or --mode=ask, this mode allows developers to query their projects, explore architectural decisions, and understand existing implementations.

Cloud Agent Handoff

Background Execution

Cursor introduced cloud agent handoff, a feature that allows developers to transfer CLI conversations to cloud agents for continued background execution. By prepending & to messages, developers signal that work should continue in the cloud rather than blocking the local terminal.

Once transferred to cloud agents, work continues autonomously in Cursor’s cloud infrastructure. Developers can close their terminals, shut down their machines, or switch contexts while agents continue executing tasks. Results become accessible via web interface at cursor.com/agents or through Cursor’s mobile applications.

Visualization Improvements

Word-Level Inline Diffs

The CLI gained word-level inline diff visualization, providing granular highlighting of exactly what changed within each line of code. Rather than highlighting entire lines as modified, the new diff view shows precise word-by-word changes.

MCP Authentication

One-Click Authentication

Cursor simplified MCP (Model Context Protocol) server authentication with a new one-click workflow. This streamlined process enables immediate agent access to authenticated external tools and data sources without complex configuration.

Additional Improvements

The release included eight additional improvements: hooks for session lifecycle customization, usage statistics accessible via /usage for monitoring token consumption and costs, WebFetch and WebSearch tools with granular approval controls, environment details via /about, enhanced message queueing performance, multi-terminal keybinding support, improved markdown rendering, and better state management during terminal resizing.