Mitmproxy 7.0 | Hacker News Reader

Mitmproxy 7.0

mitmproxy.org

258 pointsmhils75 comments

Mitmproxy is so awesome, I use it all the time! Can recommend it for anyone who needs to do some fiddling with HTTP(S) traffic for debugging / reverse engineering purposes.

I'm slightly too much of a GUI person to use it in a "explorative" phase, then I'm more a fan of Charles/Proxyman, but whenever I wanna modify the traffic or do some scripting to i.e log certain traffic, I reach for Mitmproxy.

I'd highly recommend https://httptoolkit.tech/ for that explorative GUI phase. I found it recently and the rule configuration, UI and interception setup is significantly better than Charles/Fiddler/Proxyman.

It’s such a bummer that a tool like this has a subscription model pricing. I don’t know what it’s like in other companies, but where I work getting a recurring subscription approved is a hassle compared to a one-time thing. I’ve been looking for something like this for work but a recurring cost is almost an instant dealbreaker.

Hi! I'm the dev behind HTTP Toolkit.

There is a subscription for Pro features, but most users don't need it - you can do most things common use cases need (all interception & inspection, manual traffic rewriting, almost everything that doesn't need advanced configuration) with just the free version. It's also 100% open-source, even the paid bits.

That said, imo subscriptions provide lots of benefits as a model for both sides. Basically by aligning how much value users get with how much they pay - cheap for quick use, expensive for extensive heavy use, strong incentive to fix bugs & support existing users instead of chasing the next shiny launch.

I hear your point about approvals anyway. If subscriptions are impossible, I can do one-off indefinite licenses - send me an email at tim@httptoolkit.tech

I like the look of the product, but there is no mention whatsoever of websockets in the docs.

That is the thing missing from Charles for our QA folks.

It is highly annoying to have to hook the iOS app up to Xcode just to be able to see the wss traffic when tracking down a bug.

Yep, websockets are coming later this year, but it's not done yet I'm afraid.

Tracked here: https://github.com/httptoolkit/httptoolkit/issues/36. You can subscribe to the GH issue for updates if you're interested.

I've successfully used Fiddler Classic as alternative to Charles to look at live websocket traffic when debugging things before. Websocket support in general is a bit of a hidden feature though since you need to go to the request that established the connection and double-click the websocket logo to the left on it, which opens a new panel showing its traffic.

The newer "Fiddler Everywhere" might also be able to do it, possibly even better, but I dislike the UI and I'm used to Classic so I've stuck to that.

Thank you for clarifying. I initially misread the pricing page and thought the free could only be used for personal and not commercial use. That really helps with adoption. If I can get multiple people to try it and if they like it, it’s an easier conversation with the people who approve expensed if I’m not the only one vouching for a piece of software.

I get what you mean about aligning incentives. I do want to support devs in a way that these types of projects are sustainable, but the way other products shut off access to your stuff once you stop paying make subscription models an unnecessary risk. Having a free tier like yours really helps.

While I don't currently need such a tool at work, I just want to chime in to agree with the approval point. It's significantly less hassle to get an expensive one time purchase approved than a €5 recurring payment. I very much appreciate your openness in that regard and I'll keep it in mind for the future :-)

Can confirm that the only way I could buy a subscription tool would be with my own money, they stress of buying it at work would be insane. Every monthly credit card charge would require someone to code it against the requisition, have that approved, then file in the records management system. Just purchasing software once is painful. We haven't even got to the part about installing it. You can imagine their happiness about mitmproxy.

Generally I try to buy the tool and three years of support upfront. Otherwise I'm at risk of it being deleted because we can't have software if it doesn't have a support contract 'in case there are bugs'. These rules are written by cretins in 'IT' and accountants.

Anyway, put the bit about flexible licencing on your order page.

Thanks for the kind words! We have a Google Summer of Code student working with us on mitmweb, our web interface, at the moment, so we'll hopefully have a better GUI story soon as well :)

Yeah it's really hard though these days to sniff traffic from Android because most apps choose not to trust user-added root CAs anymore :(

It's not quite that. It's usually certificate pinning. Our iOS app won't talk to anything not issued by the list of CAs we include at build time.

This is specifically to prevent mitm attacks on the app.

Yeah certificate pinning is a thing too. But since Android 7.0 apps also have to specify whether they allow user-added root CAs and most don't!

You have to get the APK and mess with the manifest to bypass that. And bypassing pinning is even more difficult.

Edit: pimterry below me explains it better: https://news.ycombinator.com/item?id=27857429

I find this a problem because MITM isn't always bad. It can be used for inspecting apps for tracking and other malicious practices by app makers.

> But since Android 7.0 apps also have to specify whether they allow user-added root CAs and most don't!

I was unaware of this as I do little interaction with Android. TIL.

And yes, I do appreciate being able to MITM all kinds of apps, especially ones that talk to a device I own to be able to add functionality/integrations not supported by the manufacturer.

I also see the benefit of preventing MITM, in our case, it is because we intentionally offload logic onto the client to simplify GDPR/CCPA compliance(can't store what you don't collect) among other design decisions which mean a malicious client could circumvent rulesets we have in place.

> which mean a malicious client could circumvent rulesets we have in place

Note that your HTTP certificate pinning is easily bypassed on both Android and iOS by a determined actor. A malicious client will circumvent rulesets you have in place. If you do not validate client inputs, you will be exploited.

Indeed, it doesn't really differ that much from DRM. It's an attempt by the author of the software to impose control over the user, preventing them from using the software how they see fit.

If I've trusted my own root, and you've decided you wont trust anything on my root, I now need to take steps to disable the certificate validation routines completely on my device, in order to bypass your restrictions.

I get that there are cases where it's extremely valuable (and in some cases, required by regulators) to enforce certificate pinning, e.g. mobile banking apps, but there are plenty of cases where it's just bullshit DRM (e.g. gym app that performs unauthenticated requests to an https endpoint to retrieve occupancy stats on each of the chain's gyms).

In the latter case, all I wanted was to track how busy the gym was over time, so I could adjust my schedule to enable me to go when the gym was quiet. I couldn't be bothered to jailbreak my iphone and install nabla-c0d3/ssl-kill-switch2, so instead I quit my gym and switched to one that was less stupid.

> A malicious client will circumvent rulesets you have in place. If you do not validate client inputs, you will be exploited.

I agree. That is why there is still serverside enforcement, but performance tuning is heavily cache based and many client request strategies are specifically designed to not bust the cache and cause a high latency call.

Also, in our case, the app is the product. I'm not very positive on it's market fit, so you may never hear of it if it launches to a fizzle, as I suspect is a large probability.

I keep a jailbroken iPhone around to bypass cert pinning. Using frida or this handy project: https://github.com/nabla-c0d3/ssl-kill-switch2

To prevent the users from accessing their own data? Any other scenario is infeasible.

I just used this app the other week do dig into what I think is a bug with rclone. Seeing the exact http traffic to / from the AWS s3 endpoint made it much simpler to understand what was going on. The CLI is super intuitive, it's a great application all around.

Mitmproxy was an interesting solution to a problem I had a few weeks ago. One of my IT support gigs was trying to use "Mevo"-branded cameras to stream some live events to Facebook using an iPad. We had to use the venue's Wi-Fi network. We found the Mevo app would not detect that we were "connected to the Internet" and wouldn't allow us to stream.

The venue's network admin said they weren't filtering anything outbound to the Internet. We could access websites from the iPad just fine. Same w/ Facebook, Youtube, etc.

I put up Mitmproxy, an adhoc Wi-Fi network on a second Wi-Fi NIC, a DHCP server, and iptables NAT on my laptop. I set the iPad to use my adhoc SSID and my machine as an explicit HTTP proxy. My intention was to snoop on the traffic to determine what Mevo was using for criteria to consider being "connected".

For whatever reason the Mevo app just started working in this configuration. I ended up sticking a spare Wi-Fi router and a PC w/ Mitmproxy installed in the venue for the duration of the event. I never did figure out what the Mevo app didn't like about the venue's network.

If I had to guess, there may have been some sort of captive portal in place. I've seen even the "invisible" kind that don't require user intervention cause interference with connectivity checks for some devices.

If I had to guess why you had issues, the app probably had certificate pinning. Apple makes it pretty easy to pin domains, but it's also not too difficult to circumvent it.

This looks to cover it pretty well: https://www.guardsquare.com/blog/leveraging-infoplist-based-...

Does anyone know how the story currently is for MITM'ing Android?

I had fun reverse engineering app APIs a few years back, but I looked into it again more recently and found that Android has started pinning certificates by default even on apps that didn't pin anything themselves. I also had trouble getting my custom certificate to be used at all, when it used to be pretty easy.

Can Frida still bypass pinning, even this new default one, or is it done at OS level?

In short:

* By default, all modern Android apps only trust CA system certificates, and on a normal device you cannot change those.

* You can change system certificates on rooted devices and most (but not all) emulators.

* Apps can opt-in to trusting user-installed CA certificates within their manifest config, but almost all don't.

If you're on a non-rooted device, the only way to MITM traffic is by modifying the application itself, so that it opts into trusting your cert. You can either inject Frida-Gadget to do that, do it manually, or use https://github.com/shroudedcode/apk-mitm which tries to do it automatically (but it's a bit hit & miss whether it works).

If you're on an emulator/rooted device, it's totally possible, see https://httptoolkit.tech/blog/intercepting-android-https/#in... for how that works.

That's separate to explicit certificate pinning. That's also possible on Android (with some standard config settings, or manually in code) although it's got much less popular I think since the defaults were tightened up. In that case you do usually need Frida, and it's generally easiest to just use a rooted device. I wrote a blog post about exactly how to do that last week: https://httptoolkit.tech/blog/frida-certificate-pinning/#how...

> By default, all modern Android apps only trust CA system certificates, and on a normal device you cannot change those.

Sure you can. It's in a settings pane, Settings -> Security -> Encryption and Credentials -> Install a certificate.

MDM profiles can also install certificates.

There's pretty much zero possibility of this being removed as a feature as it is essential to a wide variety of corporate environments.

This doesn’t bypass cert pinning in apps which was the original question

Those certificates are user-added certificates, not system certificates. Android has two separate certificate stores, the UI only exposes the user certificate store, and by default apps only trust the system certificate store.

This was changed back in 2016, the original announcement post has some good details: https://android-developers.googleblog.com/2016/07/changes-to...

Enterprises can add certificates to the system store I believe, but only as part of the initial provisioning of the device, and only for enterprise-managed devices.

Indeed, and such enterprise-management devices must be wiped for enrolment.

There's some management modes which don't require that but they don't offer the ability to install system certs.

I love you, Pim. I never thought to change the system CA store, I always thought they'd pin to their own cert anyway.

Mitmproxy dev here, happy to answer any questions! :)

I'm interested in how you generate your docs. I'd like to do something similar using Jekyll.

Our main docs are built with Hugo (https://github.com/mitmproxy/mitmproxy/tree/main/docs). For our API docs we use pdoc (https://pdoc.dev), which integrates well with most static site generators. pdoc is also maintained by us. :)

Thanks. When you began working on the main docs, did you start with a pre-built theme or did you start from scratch?

If I'm not mistaken they are built from scratch, but that part wasn't built by me. IMO the by far hardest part is actually writing the docs. :)

How practical is it (or not) to intercept TLS traffic for IoT devices on my network where I don't have the option of setting a proxy?

I've been on a crusade the past few years to try and get my self-signed CA onto "smart" devices in my home, but with some it's difficult or I risk damaging expensive hardware.

I've tried various ARP poisoning techniques and such in the past with limited success.

Many IoTs have local search and scan, so block the DNS server at your firewall then provide your own DNS server via your own DHCP server.

I think that's a classic "it depends". I see two main challenges for IoT monitoring:

1. Getting the IoT traffic to your proxy machine. As you say this is fairly trivial when you can set a proxy, but a bit tricky otherwise as you need to run mitmproxy in transparent mode (https://docs.mitmproxy.org/stable/concepts-modes/). With some luck you can set a custom gateway on your device which simplifies the setup, otherwise you need to have the capability to set iptables rules on the router. I personally have a small $20 OpenWRT router dedicated for that. Alternatively you can run your own DNS server and point everything to mitmproxy. v7 now also supports SNI/host-header based proxying! The good news is that this setup is the same for all your devices and once it works, it works.

2. Getting the device to trust your (personal) mitmproxy CA cert. This one really depends on the particular IoT device and may involve some serious reverse-engineering if you are unlucky. I wish there would be a more of a "I should be able to inspect my own traffic" movement, but it's a tricky problem to solve.

...because of enforcement on cheap Chinese etc goods.

The first part is the easy part, I have routers and run DNS, and have experience with IPTables, I can usually find ways to get hold of the traffic.

The second part is the challenge, and for some devices I've been unable to embed my own CA. I was hoping there was a proxy trick or some magic sauce I was missing. Perhaps where DHCP could automatically configure clients with a (insecure) proxy or something along those lines?

> I was hoping there was a proxy trick or some magic sauce I was missing. Perhaps where DHCP could automatically configure clients with a (insecure) proxy or something along those lines?

Nope, you're not missing anything.

Thanks for confirming, will save me some time messing around! Ill stick to getting my soldering iron out and EPPROM dumping/flashing.

You may be able to intercept a firmware update and load a binary poisoned with your own CA cert. (Lots of factors at play here, of course.)

I'm working on a similar problem (https://github.com/elahd/esp2ino/issues/16) with a project I maintain to sideload IoT device firmware (https://github.com/elahd/esp2ino).

I've been using both mitmproxy and IOXY (https://github.com/NVISOsecurity/IOXY), an intercepting proxy made specifically for MQTT. IOXY is a small, less mature project, but it's definitely worth checking out as a compliment to mitmproxy. Many devices managed via AWS IoT phone home over MQTT and, my limited experience aside, it looks like many don't bother validating certificates when authenticating over this protocol.

Thanks for mitmproxy!

Suppose I have set an interception filter and I am viewing a live packet. I am about to edit it, etc.

But when a new packet arrives, the text UI moves to the new packet! I have lost the packet that I was looking a few seconds ago.

Is this an issue in mitmproxy or am doing things the wrong way?

Could you clarify what you mean by packet? Are you talking about HTTP requests or about the new non-HTTP TCP streams? The UI/UX for the latter is very much work-in-progress.

I am talking about the HTTP requests.

The currently focused flow should never change when a new one comes in. If that's happening to you, please file a bug over on GitHub or ping me on our dev Slack and we'll get this fixed ASAP. :)

Do you have by any chance enabled focus to follow new requests? Keyboard shortcut Shift + F. Although it's disabled by default (IIRC), you may have enabled it by accident.

mitmproxy is a great tool, but it lacks upstream certificate lookup and attributes transfer to a generated host certificate. For example, Charles does that. Some applications perform attribute inspection to detect MITM proxies, namely Spotify. Are you planning to implement this feature? Or maybe you already did it in version 7?

mitmproxy does lookup the upstream certificate by default, but at the moment we only transfer common name, subject alternative names and (new) organization name. The good news is that with version 7 it's very easy to add more here, I'd be more than happy to accept contributions. :)

Good news indeed! I tried to contribute 1–2 years ago and quickly realized that such a feature would require a substantial rewrite after a quick code inspection.

If you (or anyone else reading this) is interested, please do reach out on GitHub or our dev Slack. Happy to help you get started. :)

Yes, exactly!

I am still using version 2.0 after the big changes in shortcuts. I used mitmproxy weekly, now I have switched to mostly using Charles, but would love to switch back.

Any plans for a recommended shortcut setup again?

Not writing 30 characters for copy body of a POST request, but fast key bindings.

I don't recall what changed with v3, but sorry for letting you down. Shortcuts are fully configurable since a few releases: https://docs.mitmproxy.org/stable/concepts-commands/#custom-....

If you have specific feedback about where the keyboard UX sucks I'd be happy if you could open an issue on GitHub so that we can take a closer look!

Thanks for Mitmproxy! It would be amazing to be able to see non-HTTPS (but still TLS) traffic in the Flows view, and even to intercept it live instead of having to drop to raw TCP mode and write Python. Any idea whether that's on the roadmap?

Starting with this release you do see TCP-over-TLS in the flows view, we have a small screenshot in the blog post. Interception/modification is still something where we need to iron out the UX for, with version 7 you can at least inject messages with the `inject.tcp`/`inject.websocket` commands. This is not very well documented yet, but the technical foundations are there now. :)

If I wanted to use mitmproxy to mitm and observe traffic during automated testing, what mode/settings should I be using on Windows?

The setup is: A client (dev machine or pipeline) running tests that communicate to a Windows VM that hosts the app under test. The client RPCs to call automation APIs on the host/server.

However I couldn't get both communicating and using mitmproxy to observe traffic to work at the same time.

If your server is really just one service on one port, I would just run mitmproxy as a reverse proxy and point your client to that (https://docs.mitmproxy.org/stable/concepts-modes/). If there are multiple services, having it running as a regular proxy sounds best to me.

I know that's not the purpose of mitmproxy, but wondering if it's possible to create automation scripts, like for puppeteer or export requests to curl/httpie/vscode rest client. I tried through web-interface but couldn't find anything like.

You can export to cURL/httpie export in mitmproxy, it'll be exposed in mitmweb with the next release. :) For automation, take a look at https://docs.mitmproxy.org/stable/overview-features/#server-...!

Is anyone using this for page rewriting? Not just for the all-important ad blocking but removing the whitespace where the removed ad was, rewriting autoplay videos and gifs into click-to-play, etc. So many apps have web views that the doing it in the browser is a step backwards.

I used to do all my rewrite via a proxy but it stopped being particularly useful when https became almost universal.

It's getting harder and harder to do even with HTTPS! Look at the other commenter on here that described the absolute wasteland with apps and OS-vendors (Android,IOS) being actively hostile to device-owners.

Edit. Typo.

Such an amazing and well-maintained tool!!!

Anyone interested in sharing use-cases where it has helped you?

Scripting flow manipulations in mitmproxy is easy. There are a lot of great examples to learn from which help. The filter syntax is expressive and simplifies writing code to mangle flows. I've done body conversion [0], weird types of authorization testing, manipulated system update packages on the fly, all sorts of bizarre things. Auto-reloading scripts on change is such a time saver. If you want a python-based programmable HTTPS proxy for testing I have not found anything better.

[0]: https://carvesystems.com/news/proxying-wbxml-services/

We have a super diverse userbase, so this is a fun question! Some pointers:

Development:

- Debug your own apps/devices (which do not come with Chrome Devtools).

- Simple Python scripts to replace/inject website content. (https://docs.mitmproxy.org/stable/addons-overview/)

- Capture a web page/API and then replay the responses offline. (https://docs.mitmproxy.org/stable/overview-features/#server-...).

- Replay requests for testing. (https://docs.mitmproxy.org/stable/overview-features/#client-...)

Privacy:

- Figure out what your Covid app is sending out (https://seancoates.com/blogs/how-i-helped-fix-canadas-covid-...)

- Look at how evil IoT devices for pets are (https://www.nytimes.com/2018/12/11/realestate/spying-on-the-...)

- Generally find more evil stuff (https://mitmproxy.org/publications/)

Security:

- Reverse-engineering custom binary protocols is never fun, but with mitmproxy you can at least write custom contentviews to dissect stuff. (https://github.com/mitmproxy/mitmproxy/blob/main/examples/ad...)

- Fiddle with generic TLS-protected TCP streams. (https://github.com/mitmproxy/mitmproxy/blob/main/examples/ad...)

- Dump TLS master secrets for WireShark. (https://docs.mitmproxy.org/stable/howto-wireshark-tls/)

We have a product that depends on data received from external web services. We often want to test certain conditions in the data (specific results, error scenarios, etc), and it's difficult or impossible to guarantee these conditions when we call out to our external services. We use mitmproxy to intercept and manipulate the responses we get.

Even if we don't need to manipulate the data, it's a very useful tool for watching requests go back and forth.

I’ve used it to diagnose (and work around) webhook problems between third-party applications that are supposedly compatible.

One that sticks out in my mind is trying to receive Slack webhooks in Mattermost. I was unaware of Slack’s new BlockKit message format until I saw what was being sent. Then I wrote a simple python script to transform the payload, all within mitmproxy. It’s not the most efficient long term solution, but the fact I could diagnose and resolve the issue with a single tool, all within a few hours of first discovering it, was very satisfying.

Here’s the result for anyone interested (warning: it’sa quick hack, not a polished/maintained project)

https://github.com/thinkmassive/mattermost-webhook-slack-pro...

Edit: this doesn’t use TLS at all. I understand that’s probably the primary use case for the tool, but I thought this was a cool example of other ways it can be useful too.

Mattermost CEO here, really nice work @thinkmassive!

I've introduced mitmproxy in my team and we use it quite often for flows' inspection and manipulation while testing a desktop app. Can't count cases where a quick look into a proxy made debugging a lot easier, because we were able to point at unexpected requests / responses.

Thank you for your work, @mhils & the team. Enjoy your moments of glory on HN :)

I use it to download music from my music streaming subscription to my son's iPod. None of the files have any tags or metadata so it's was an interesting challenge to get everything tagged.

I wrote a krunkerio aimbot that unpacks the msgpack-over-websocket packets and injects the calculated mouse movements into the stream.

The biggest problem was the server only syncing the player position every second or so which means my middleware has can only provide exact calculations when I stand still. Solving this would require reading the process memory but that is exactly what I wanted to avoid :/

I’ve used this tool to identify and fix security issues and have also taught software testers how to use it just to spread the love and knowledge. Kudos!

That sans io core sure sounds like a free monad interpreter.

Reply on news.ycombinator.com