Image: GitHub
Why it mattersA team lead can see one month of AI spend without asking engineers. The local view needs no account and sends no code off the machine.
A team pays for Claude Code, Cursor, ChatGPT and Gemini and nobody knows who spent what. TraxAgent has released a Claude Code plugin that reads the session logs on the laptop and answers that one question.
What the plugin does
The project is an MCP server and a Claude Code plugin. It adds a /spend-audit command and a /mcp command, per the README. The local command reads Claude Code's own session logs on the machine. It then estimates spend using public API prices. The README says the command needs no account and no server connection.
The /mcp command connects to TraxAgent's own read-only MCP server after an OAuth sign-in. That path covers more services. The README lists Anthropic, OpenAI, Gemini, Claude Code, Codex, Cursor and AWS Bedrock seats. Both paths are MIT licensed.
The repository reports 137 stars and 31 commits on its main branch.
What a team sees
The local audit answers a simple question: which Claude Code sessions cost the most on this laptop, and which model did each one use. The remote server adds the team view: who in the team spent the most, which model each person uses, which sessions ran at a high per-run cost. TraxAgent writes the recommendations as ranked actions with a dollar range beside each one.
The README shows a worked example. It describes a sample company with 39,000 dollars of monthly AI spend. In that example, TraxAgent reports 9,200 to 18,700 dollars a month is avoidable. One recommendation shifts work to the Batch API for a 2,800 to 4,000 dollar saving a month. Every figure is TraxAgent's own, from its sample company.
Where the demo stops
TraxAgent also runs three tools at traxagent.io that need no sign-in. A free audit takes a CSV file of usage. A demo dashboard shows the sample company data. A model speed comparison ranks models by cost per successful run.
Two cautions sit around the piece. The sample company numbers are illustrative, not measured on a real team. And the richer view only arrives through the hosted server, which reads a team's billing data with admin keys. A team that pays for AI across four or five services has a quick path. It can run /spend-audit and stop there. A team that wants the full report hands TraxAgent read-only billing access.
The broader pattern is that AI spend has become a line a manager can see and act on. Apple cut its engineers' AI daily budget from 300 to 200 dollars this month, as Bloomberg reported. A team that does not yet know who spent what has no way to make the same call.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

