Start coding in Cursor, finish in Claude – full context follows you (github.com)

Cursor now preserves and hands off your full coding context to Anthropic’s Claude, letting you start work in Cursor and continue in Claude without losing state. The release (noted in the v0.4.0→v0.5.0 changelog) emphasizes continuity — capturing editor state, open files, recent edits and related metadata — and exposes qualifiers to control what gets shared (see the docs). The announcement also highlights user feedback shaping the feature and points to standard error-handling behavior (reload on load errors) and updated documentation for available qualifiers. This matters for AI/ML practitioners because it reduces friction between an IDE-like code-first workflow and large-model-assisted coding or code review: you can migrate to a conversational model with context intact, preserving reproducibility, prompt fidelity and the local project scope that models need to generate accurate fixes or design suggestions. Technically, it implies richer context windows and tooling integrations (with explicit qualifiers to limit scope), raising practical implications for prompt engineering, data privacy, and how models consume repository state. Overall, the update signals tighter toolchain integration that streamlines model-assisted development while giving users controls over what context is shared.