See the tokens Claude Code and Codex load before you type: MCP tool lists, skills, hooks, CLAUDE.md and AGENTS.md per session. Find what you never use and archive it, with undo. Open source, runs offl
See what your coding agent carries, keep what you use, archive the rest.
Claude Code loads every skill, agent, hook and MCP server you have installed into every session, used or not. packlight scans your setup and your local session logs, then writes one self-contained report. The report shows:
- how many tokens each session starts with, by source: MCP tool names, the skill listing, hook text, agent descriptions - which skills are pushed out of Claude Code's skill listing - how much text each hook injects - which items were not used in any session since they were installed
Works with Claude Code and Codex. packlight picks Claude Code when ~/.claude exists; for Codex, add --agent codex (packlight --agent codex, packlight fix --agent codex). Each tool gets its own report: ~/.packlight/report-claude-code.html for Claude Code, ~/.packlight/report-codex.html for Codex. For Codex the headline is measured: Codex logs the input tokens of each session's first request, and packlight reports the median. Codex plugins and MCP servers are switched off in ~/.codex/config.toml, so packlight lists them with the exact line to change rather than editing that file.
See a sample report → (an invented setup, so you can click around before installing).
You mark what to archive. packlight apply moves it out reversibly, and packlight restore brings it back.
Everything runs locally: no telemetry, and no network calls from the CLI or the report.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/cooperism-lab--packlight.html)
Keep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
magna-nz/tallybookPrices your Claude Code and Codex sessions from the transcripts on disk, shows what each agent cost over any period, and tells you in plain English what would have been cheaper. CLI, Browser UI and MC
azimov777/casefileThe task tracker your AI agents keep for each other: every task carries a case file, so the next agent picks up where the last one stopped. MCP server + live board. Self-hosted, one-line install.
00200200/tokencutKeep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
cuongnbms/mastering-claude-codeCẩm nang thực chiến Mastering Claude Code (tiếng Việt): 16 chương biên soạn từ tài liệu chính thức của Anthropic — CLI, workflows, MCP, subagents, skills, hooks, plugins, CI/CD & tối ưu context.
0xwilliamortiz/openclaude-improvedruns anywhere. uses anything
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.