Skip to main content
Tracks tells you which coding agent gets your tasks right, and what it costs. Give Claude Code, Codex, Cursor, Claude Cowork or the Codex app the same tasks, with or without your MCP server or skill, and compare how often each one succeeds, what it spent and whether it used your tools. Each agent works in its own cloud sandbox, and Tracks records everything it does.

​
Start here

In the web app

Compare two agents on your first task. No terminal needed.

From your agent

Paste one prompt into Claude Code or Cursor and let it set Tracks up.

With the CLI or API

Start a run from your terminal, your code or your CI.

​
What you can do

  • Run an agent on a prompt, on a repository or none, and read what it did, step by step, with what it cost. See runs.
  • Compare agents with an experiment: the same scenarios given to several agents, each run judged on your own yes-or-no evaluations. See experiments.
  • Read the results: the pass rate per agent and per set of tools, the total spend, and how many runs used each of your skills, MCP servers and command-line tools. See reading results.
  • Share a run with a public link, and invite members to your organization. See organization and sharing.
  • Pay as you go from a prepaid balance. See billing.
The web app also has Analytics, which shows who visits your website, AI agents included, once you connect Cloudflare or Vercel.