ClaudeRabbit
Sign inStart scanning
It is live, free and open source.

ClaudeRabbit

ClaudeRabbit is a security audit tool that scans open-source repositories and packages to protect the developer and builder community from malware.

AIdhirajSingh/cr-exfil-test-fixture00:00.0Booting

A replay of the sandbox run on AIdhirajSingh/cr-exfil-test-fixture: Malicious, 0 out of 100.

GitHub holds 395 million public repositories, and GitHub is only one platform.

There is also npm, on every developer's laptop, because every piece of software is built from packages. Open source moves overnight.

A repository can get 100 thousand stars in a day, which means 100 thousand people ran it.

395M
public repositories
100K
stars in a day
450,000
new malicious open-source packages

Who did its security audit?

Take OpenClaw, one of the most viral projects on GitHub.

It takes your computer's permissions and your AI API keys. It was technically invasive, with a real chance of harm, and people still ran it, trusting open source.

In 2025 alone, over 450,000 new malicious open-source packages appeared.

Paste a GitHub link or an npm package.

ClaudeRabbit boots a fresh, disposable virtual machine, downloads the code, installs it and runs it, the same way you would. Every outbound call it makes is answered by our forge with a fake reply and recorded, so nothing real leaves the machine. Static scanners read the code in parallel. Jev, a fast decision model, answers the security questions. A deterministic formula in code turns that evidence into the score. Luna writes the report in plain English. The machine is destroyed after every scan.

Fresh, disposable virtual machine

Downloads, installs and runs

Our forge answers with a fake reply

Static scanners read the code

The score, then the report

The machine is destroyed

The score is well calculated.

It shows the evidence behind every point it takes off, and the owner's reputation is shown separately. ClaudeRabbit never just says "safe". It tells you what it ran and what it found with a proper safety score.

lodash/lodashNo threats observed
90/ 100
Observed when runlow
guarddog: guarddog.capability-network-outboundlow

It works on the web, in your terminal through the CLI, and inside AI agents through MCP.

github.com/expressjs/expressScan
90No threats observed

If every repository got this audit the moment it launched, developers could stop worrying about security and focus on what the product does.

It helps the people who run code and the people who publish it.

$0.001

A deep scan, with the server, the AI, the web, the sandbox and the network included, costs us about $0.001.

That is what makes it free for everyone, everywhere. ClaudeRabbit runs on programmatic advertising. All it needs is your support and awareness.

Cybersecurity for builders.

Start scanning