TokenLens — agent context diet dashboard

Scan local coding agent configs and skills to reveal what is quietly burning tokens.

Problem

More people complain their quota runs out but cannot pinpoint which SKILL, MCP config, or memory file is eating tokens. Some tried RTK truncation only to be told key info was dropped, leaving them to manually tweak CLAUDE.md, skills, MCP, and hooks by gut feel.

Solution

A local desktop tool that reads each coding agent’s config and skill directories, estimates fixed context cost per skill and MCP, and charts token usage per session. It surfaces concrete advice like disabling a skill, switching to manual trigger, or using an index, with one-click backup and config switching.

Users

Heavy developers using Claude Code, Codex, or Cursor daily with tight quotas

MVP

Scan local agent configs and skill folders; rank fixed-context cost; parse session logs into token curves; give actionable slimming suggestions; one-click backup of current config

Revenue

Free tier for rankings; Pro subscription for auto-suggestions, config comparison, and shared team templates

Why now

Token cost is now a major monthly line item for developers, yet tooling only offers manual trial and error with no quantified view