Token Counter

Token Counter is a plugin for Codex and Claude Code, published through the tokenusage.dev site. It reads the usage data those tools record on your computer and builds a local report showing where your tokens went. According to the site, none of your usage leaves your computer unless you choose to share it.

What It Is

Token Counter is a usage-reporting plugin for two coding agents, OpenAI's Codex and Anthropic's Claude Code. Users install it from a plugin marketplace and then ask the agent in plain language, for example "Generate a token report", to produce a local report. A separate request, "Share my token usage to tokenusage.dev", publishes the report to a shareable page. The site links a sample Codex report.

What the report shows

• How each weekly limit was spent, comparing the limit percentage the tool recorded with running usage for that week, restarting at every reset.
• Tokens used each day, so the heaviest days stand out.
• Input by model, and the session that used the most.
• What filled the context: tool output, your messages, instructions and other content.
• The days a limit was recorded as reached, beside each day's response times.
• The same tokens priced at API list prices.

The Codex report counts context in tokens. The Claude Code plugin ships no tokenizer, so its report measures context in bytes of visible text instead.

Setup path

For Codex CLI, add the marketplace with codex plugin marketplace add and install the plugin. In the Codex ChatGPT desktop app, add the marketplace, restart, and install Token Counter from the Plugins screen. For Claude Code, the separate token-counter-claude plugin comes from the same marketplace, reads Claude Code's transcripts and retained capture history, and installs no packages. Its report, capture history and share token are separate from the Codex plugin's, and a new session is needed after installing. Claude Code also offers slash commands for the local report and for sharing.

Related tokenusage.dev resources

The surrounding site also publishes comparisons of Codex and Claude plan limits, weekly limit estimates, response-time and output-speed charts, and reset tracking.