Guide

AgentLens shows what coding agents cost in tokens: which skills, tools and instruction files fill the context, which of them nobody uses, and what a month of sessions costs. It reads the files that Claude Code, OpenCode, Copilot CLI and VS Code Copilot leave on your machine, and it comes as a command line, a desktop application and an MCP server for agents.

The pages are listed in reading order; the list is also the sidebar of the site.

1. Download: the application, the command line and the MCP server, and how to check them.
2. Getting started: install, first commands, where the numbers come from.
3. Sources: which agents are read, where their files are, and what each one reports.
4. Measured and estimated: how to read a number, and why some are guesses.
5. The analyses: skills, context, unused, flow, cost, efficiency and compare.
6. Reports and export: JSON, CSV and the offline HTML report.
7. Compatibility: what in the output is stable and what may change.
8. The desktop application: views, filters, export, updates.
9. The MCP server: let an agent ask AgentLens itself.
10. Exact token counts: the one opt-in feature that sends text away.
11. Privacy: what is read, what is written, what leaves the machine.
12. Glossary: the terms in one place, the same entries as the application.
13. Importing OpenTelemetry data: exports of Claude Code, Copilot CLI and VS Code Copilot.
14. Updating the application: how the update check and the installation work.