Cursor 3.3 adds agent context breakdown to debug token use

Cursor has just rolled out a new agent context usage view in Cursor 3.3, aimed at helping users diagnose context issues across rules, skills, MCPs, and subagents. Users are already asking for deeper breakdowns by project, tool, and CLI access.

cursor cover

TL;DR

  • Cursor 3.3 context breakdown view: Shows how agent context is used; helps diagnose context issues
  • Setup tuning: Aims to support adjustments across rules, skills, MCPs, and subagents
  • Requests for monthly usage visibility and per-project breakdowns
  • Requests to surface stats in Composer and printable output from the CLI
  • Need for itemized context contents: which tools, rules, skills are added, not only totals
  • Cursor developer: more visibility features and context-optimization tools “coming soon”

Cursor has added a breakdown of agent context usage in Cursor 3.3, with the company posting that the new view can help users “diagnose context issues” and adjust setups across rules, skills, MCPs and subagents.

The post drew a steady stream of replies asking for broader visibility. Some users requested an overall monthly usage breakdown, while others asked whether similar stats could be exposed in Composer, printed from the CLI, or broken down by project. One commenter also pointed to the need to see which tools, rules and skills are being added to context, rather than only seeing totals.

A Cursor developer replied that “lots more [are] coming soon” to provide “even more visibility + tools to optimize context.” Other responses praised the feature as a way to make agent behavior less opaque, with several users describing it as a useful debugging aid for tracking where tokens are going and why agents may behave unpredictably during longer tasks.

Source: Cursor

Continue the conversation on Slack

Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.

Join our community