Context Usage Breakdown

Source: Cursor Date: 2026-05-06 URL: https://cursor.com/changelog/05-06-26

Summary

Cursor’s May 6 changelog (v3.3) introduced a Context Usage Breakdown panel for agents: a per-component view showing how context tokens are allocated across Rules, Skills, MCPs, and Subagents within a running agent session. The feature is framed as a diagnostic tool for identifying bloated context configurations and optimizing agent setup. It sits adjacent to two other recent changes — PR Review/Build Plan improvements and upcoming model controls with spend management.

Implications

• Context visibility is becoming a first-class UX concern in agent IDEs: as agents accumulate Rules, MCP servers, and subagent invocations, token budgets become opaque and expensive. A breakdown panel is a direct response to users hitting context limits without knowing why.
• The breakdown across Rules/Skills/MCPs/Subagents mirrors the architectural layers of modern agent configurations — this taxonomy is practically a schema for how Cursor models agent composition internally, and it’s likely to influence how other agent frameworks surface context accounting.
• Feeds the agent observability thread: alongside OTel metrics in Claude Code and spend tracking in Cursor’s next changelog entry, this signals that cost and context transparency are consolidating into core product requirements, not afterthoughts.