CARBONby testers.ai

FREE. OPEN SOURCE. INSIDE YOUR CODING AGENT.

Your AI writes code.
Jay helps you
trust it.

Give your coding agent a test manager. CARBON finds risks, runs checks and helps turn bugs into regression tests—with evidence you can inspect.

Install CARBON free

Claude Code · Codex · Cursor
No separate CARBON account. Your agent’s model costs still apply.

YOUR PROJECT / CARBON
Test my latest changes.
Show me what could break.
Jay
JayYour AI test manager
  1. 01
    Follow the riskRead the change. Find the vulnerable flows.
  2. 02
    Test what mattersRun checks. Explore. Reproduce failures.
  3. 03
    Keep the evidenceReview findings, regression tests and gaps.

Workflow preview · not live results

BEFORE YOUR USERS FIND IT.

Find the bug.
Keep the test.

Ask CARBON to challenge your latest AI-generated change. Reproduce a failure, inspect the evidence, and turn what you learn into a regression test.

Find my first bug →

Free and open source. Inside Claude Code, Codex or Cursor.
Your agent’s model costs still apply. Finding a bug isn’t guaranteed.

ONE CHANGE. THREE QUESTIONS.

What could break?

Follow the change into the flows your users depend on.

Can we reproduce it?

Separate suspected problems from failures backed by evidence.

How do we keep it fixed?

Propose a regression test in the framework you already use.

Illustrative workflow · not live results

Jay, CARBON’s AI test manager

MEET YOUR CODING AGENT’S TEST MANAGER.

Your agent builds.
Jay challenges it.

Give testing a seat inside your development workflow. Jay uses your project context to prioritize risks, run available checks and explain what’s still uncertain.

Add Jay to my coding agent →

CARBON is free and open source. Claude Code · Codex · Cursor.
No separate account. Your agent’s model costs still apply.

Risk analysisExploratory checksRegression testsEvidence

START WITH SOMETHING YOU CAN TEST.

Less promise.
More evidence.

Try CARBON on a disposable demo before using it on your own project. Give Jay a flow to test, then inspect the findings and coverage gaps.

Try the free demo workflow →

Install in Claude Code, Codex or Cursor first.
Node.js 22+ and Python 3.10+. Your agent’s model costs still apply.

YOUR FIRST REQUEST

“Create a demo.
Test one important flow.
Show me the evidence.”

Start small. Inspect the results. Then bring the same workflow to your own code.

Get set up and copy the demo request →

Workflow illustration · not a live or pre-recorded run

Help improve this page? Optional Fathom measurement

If you allow it, Fathom measures the page version and fixed-name actions such as clicks and copied test requests. We remember your page version for up to 30 days, without creating a visitor ID. No email, code or report content is included in event names. Clicks are not verified installs or sign-ups. Your choice won’t affect installation. Details

Measurement off.

ONE PLUGIN. YOUR EXISTING WORKFLOW.

Choose your coding agent.

Install CARBON. Open your project. Give Jay one thing to test.

Add CARBON to Claude Code.

  1. Open /plugin and choose Discover.
  2. Search CARBON. Check that the publisher is testers.ai, then install.
  3. Start a fresh session and choose /carbon:carbon.

In Claude Cowork, use the plugin directory. Your organization may restrict plugins.

Installation help →

Add CARBON to Codex.

Open the listing, verify the testers.ai publisher, and install in your supported local Codex environment. Start a fresh session and choose CARBON in the skill picker.

Open CARBON in the marketplace ↗

The local plugin needs your coding agent’s tools. Ordinary ChatGPT web chat is not the local Codex runtime.

Connect CARBON through MCP.

In Cursor’s MCP settings, add a local stdio server using this configuration. Reconnect, then ask your agent to use CARBON.

{
  "mcpServers": {
    "carbon": {
      "command": "npx",
      "args": ["-y", "@testers-ai/carbon-mcp"]
    }
  }
}

Node.js 22+. Some Windows hosts need npx.cmd. Other MCP hosts may use a different configuration wrapper.

Package & setup details ↗

More than “the tests passed.”

Test the change, explore the experience, and see what remains uncertain. Keep your findings. Keep your existing framework.

A few things worth knowing.

Is the useful version really free?

Yes. The core CARBON harness is MIT-licensed and open source, including the comprehensive /full workflow within its included capabilities. Pro adds specialist controls and advanced workflows. Your coding agent’s subscription and model usage are separate.

Does it replace my test framework?

No. Jay works through the tools available to your coding agent. Keep Playwright, Selenium, Cypress, pytest, JUnit or your current runner. Tests still need the right dependencies and access; a generated test is not a passed test.

Does it test everything automatically?

No. Scope, edition, host permissions and budgets matter. MCP and public Codex-directory builds run on explicit requests. Native builds with supported hooks offer a saved testing-mode choice. Review settings before enabling automatic checks.

Where does my code go?

CARBON runs through your coding agent’s environment and tools. Your host’s model and data-handling policies still apply. Reports are local; review exports before sharing. Optional usage analytics and milestone measurement are separate settings. Never publish a protected report URL.

What do I need installed?

A supported local coding-agent environment and Node.js 22 or newer. Demo and full-workflow helpers also need Python 3.10 or newer. Your host supplies model access, terminal tools and browser capabilities; CARBON does not supply cloud VMs.

OPTIONAL. NEVER REQUIRED TO INSTALL.

Keep up with CARBON.

Get occasional workflow tips and release updates from Jason. Free to join. No paid account.

Request CARBON updates

Your name, email and permission are saved in Testers.ai’s private Google Sheet and emailed to Jason. Signup works with analytics off. No automatic welcome email. Privacy details