DEV Community

Cover image for I built a free, open-source Burp Suite alternative with an AI agent inside
evilboyajay
evilboyajay

Posted on Edited on

I built a free, open-source Burp Suite alternative with an AI agent inside

If you've done any web security testing, you've used Burp Suite. It's the standard for good reason. But if you're on the free Community Edition, you know the friction: Intruder is throttled to a crawl, there's no Collaborator, and you can't save a project. Pro fixes all of that for $499 a year, which is a lot when you're learning or doing bug bounties on the side.

So I built ProxyAudit: a free, MIT-licensed intercepting proxy for macOS that covers the everyday Burp workflow and adds something Burp doesn't have built in, an AI agent that can read your captured traffic with you.

The agent writing and running a proof of concept for an IDOR. Watch the full 100-second demo.

What's in it

The goal was simple: every tool I reach for in a normal testing session, with nothing throttled or locked behind a license.

  • Proxy and HTTP history: HTTPS interception with a CA generated on your machine, filtering, scope rules, and JSON/HAR export.
  • Intercept: hold requests in flight, edit them, forward or drop.
  • Repeater: edit and resend requests across as many tabs as you like.
  • Intruder: Sniper, Battering ram, Pitchfork and Cluster bomb. You choose the thread count, up to 1000 requests in parallel, instead of being held to about one per second.
  • Collaborator: an out-of-band listener over a cloudflared tunnel, for blind SSRF, XSS and XXE.
  • Sequencer, Decoder, Comparer: token entropy, stackable encoders (including JWT), and line or word diffs.
  • WebSockets, site map, content discovery, match and replace.
  • Plugins: JavaScript plugins that hook traffic, in the spirit of Burp's Extender.
  • Mobile interception: expose the proxy on your LAN and install the CA on iOS or Android from a URL.

Collaborator catching an out-of-band callback through a cloudflared tunnel

Collaborator catching a blind SSRF callback: the payload URL goes into a parameter, and the server's request shows up here.

One feature I care about for beginners: Open browser launches Chrome, Brave or Edge in an isolated profile that already uses the proxy and trusts its certificate. Certificate setup is where most people get stuck on day one, and this skips it entirely.

The AI part

ProxyAudit uses the Claude Agent SDK. If you've already run claude login for Claude Code, it uses that session. No separate API key needed.

There are two ways to use it.

One-click audits. Select a request and click Audit endpoint, or select a JavaScript file and click Audit JS. The agent looks for IDOR, injection, broken authentication, leaked secrets and dangerous DOM sinks, and writes up what it finds. You can also audit every captured JS file in one go.

Agent chat. You describe what you want in plain English, and the agent works through the proxy using a small set of tools:

  • list_flows, search_flows and get_flow to read your HTTP history
  • send_request to replay and modify requests
  • list_js_files and crawl_site to map the app
  • enum_subdomains for passive recon from certificate transparency logs
  • browser_navigate, browser_click and browser_read_page to drive a test browser

Big targets can be split across parallel recon and security-auditor subagents.

A real example: finding an IDOR

Here's the flow from the demo above, against a deliberately vulnerable app running locally.

  1. Log in as Alice and browse. ProxyAudit captures GET /api/users/1 with Alice's bearer token.
  2. Send it to Repeater, change the ID to 2, and resend. The response comes back 200 OK with Bob's email, role and card digits. That's an IDOR.

Repeater: changing the user ID from 1 to 2 returns Bob's admin record

  1. Switch to the agent and ask it to audit the captured traffic and write a proof of concept.

The agent read the history, wrote idor_poc.py, ran it, and reported back:

[+] logged in as alice (id=1)
[own  ] /api/users/1 -> 200 {"id": 1, "name": "Alice", ...}
[IDOR!] /api/users/2 -> 200 {"id": 2, "name": "Bob", "role": "admin", ...}
[deny ] /api/users/3 -> 404

1 foreign record(s) exposed
Enter fullscreen mode Exit fullscreen mode

It also pointed out the likely fix (check that the token's subject matches the requested ID, or serve /api/me instead) and flagged that the token's signature was a placeholder string, so the server might not be verifying JWTs at all. That's worth a second look of its own.

What it isn't: the agent doesn't replace knowing what you're doing. It's fast at reading a lot of traffic and at the boring parts (writing the PoC, checking every ID in a range, grepping minified JS for keys). Treat its findings as leads to verify, not as a report to paste into a bug bounty submission.

How it's built

  • Electron and plain JavaScript. No framework in the renderer, and no JVM, so it starts in a couple of seconds.
  • http-mitm-proxy for the MITM engine, with a CA generated per machine on first launch and never shipped in the repo.
  • A sandboxed renderer (contextIsolation on, nodeIntegration off) that talks to the main process only through a whitelisted bridge.
  • GitHub Actions builds Apple Silicon and Intel DMGs and publishes a release on every push to main.

What's missing

To be upfront about the gaps:

  • macOS only for now. Windows and Linux builds are the top roadmap item.
  • HTTP/1.1 and WebSockets only, no HTTP/2 yet.
  • No active scanner. The agent audits on request, but there's no automated crawl-and-scan like Burp Pro.
  • History isn't saved between sessions yet. You can export to HAR or JSON.
  • The app isn't notarized by Apple yet, so the first launch needs a click in System Settings.

Try it

If you try it, I'd love to hear what breaks and what's missing. The roadmap discussion is the best place to say what should come next, and PRs are very welcome, especially from anyone who wants to take on the Windows or Linux build.

Only use ProxyAudit on systems you own or have written permission to test.

Top comments (0)