Agentic testing,from your terminal.

One command installs it; testsprite setup hands it to your coding agent. From there, end-to-end tests run from a shell, a script, or a CI step.

Open source on GitHub

Features

Your agent verifies its own work.

One command, whether you type it or your agent does.

Terminal-native

End-to-end tests without opening a browser or an IDE — and testsprite setup teaches your coding agent the commands, the exit codes, and where a failure lands.

  • One npm install — no browsers to install
  • Claude Code, Codex, Cursor, Cline, Antigravity
  • The same verifier as MCP, CI, and the dashboard

Failures you can act on

One command pulls the whole failure bundle into your repo: the failing step, a video, a DOM snapshot, the root cause, and a suggested fix.

  • Reproduction steps your agent can follow
  • Session video for every failure
  • A fix verified by the same test that failed

A suite that compounds

Passing tests are kept rather than regenerated, so one command re-checks everything you have verified so far — and coverage grows with the app.

  • Re-runs the whole suite, not a sample
  • No selectors to hand-maintain
  • Regression coverage compounds instead of decaying

Pipelines

Scriptable by design.

A stable --output json from every command and documented exit codes, so a pipeline can gate a merge on the result.

Works with the coding agents you already use.

Plug in through the open-source CLI, or MCP in your IDE. Your agent writes the code, TestSprite verifies it, and nothing about your workflow changes.

  • Claude Code
  • Codex
  • Cursor
  • Antigravity
  • VS Code
  • Kimi
  • Trae

Apache 2.0 on GitHub — the same CLI our own agents run. Read it before you run it.

View the repository

The public record · codercup.ai

When frontier agents go head-to-head,we’re the referee.

Claude Code, OpenAI Codex and Google Antigravity build the same ten-phase app under one clock. We score every phase against 16–22 end-to-end test plans.

See the public record
0% 80%
Self-evolve
Same model, ten rounds of reading failures and fixing them.
~12%-25%
Regression
Passing features broken by a single change.
42% 92%
Requirements met
Feature delivery once a verification loop is in place.
KimiClaude CodeAntigravityCodex

Good to know

Frequently askedquestions

Node.js 20 or newer, and a TestSprite account for the API key. Run npm i -g @testsprite/testsprite-cli, then testsprite setup — it prompts for the key, verifies it, and installs a verification skill for your coding agent. If you would rather not install globally, npx @testsprite/testsprite-cli works for one-off runs.

Install it now.Verify on your next commit.

Free to start. Node.js 20 or newer is the only prerequisite.

Open source on GitHub