The disclaimer at the end of the article says AI was used to "polish" the writing. I would guess it's this polish that can be a little off-putting: It reads like a magazine article written by and for people who aren't particularly interested in the topic.
My lacklustre writing skills aside, that garbage needs to be called what it is.
It is either satire designed to grab engagement by shocking people, or the genuine anti-AI frustration of a mediocre writer who has realised they've lost their edge, because their work isn't worth paying for over AI.
The people winning literary awards will keep winning them with their manual writing, and they will keep making good money ghostwriting.
But my article, which is meant to communicate some software features in decent, readable prose, isn't worth $10k of ghostwriting. AI will have to do. So you go get, sir, and please stop gatekeeping! (Not you specifically, you get my drift.)
The point of the writing is thinking what you are about to say from difference perspectives. Now, everything gets average and boring if LLMs are getting used all the time. And they still waste so many words that don't benefit reader in any way.
I can bet author didn't even read his own AI generated text.... I don't have any issue with AI, but it would be nice if people spend time over creating quality stuff.
since GLM 5.2 (perhaps even earlier?) it has been remarkably easy to reverse engineer any closed protocol you want as long as you have a binary. previously, it required so much manual effort as to simply not be worth it 95% of the time.
now all you need is IDA or Ghidra MCP, a binary and some vague sloppy instructions.
some more recent models even started instrumenting a running binary (when possible) to enumerate the protocol without being explicitly instructed to, which is even better.
You don't even need an MCP, I just let the agent write scripts for headless ghidra. The MCPs are fragile and there isn't a whole lot of knowledge about how to use them in the training datasets, whereas there are plenty of ghidra scripts (and proper docs for the APIs).
Seriously, though, the popular Ghidra MCP is really badly architected; it's way better to rearchitect it or just script Ghidra directly. With that said, Opus 5.5 seems to have been trained on CoT from the popular Ghidra MCP. This makes it work better, but also makes it even more inefficient if you modify the MCP without changing its name and shape significantly (it will try to make tool calls using the "mainline" format, then have to retry them when they fail).
Even with Opus 5.5, IMO it's better to just ditch the MCP and let the LLMs eat with bintools and headless Ghidra; with both GLM and Opus this produces significantly more efficient results than the popular MCP. On the other hand the IDA Pro MCP is much better architected and seems to be pretty good.
> there isn't a whole lot of knowledge about how to use them in the training datasets
if you include a SKILL.md for the MCP (or just dump it into the prompt) it's not a problem.
I don't use Ghidra but IDA Pro MCP works extremely well for me for all manner of tasks. for example, some software likes to call home for license checking (and I wish to run it with networking denied to it). it no longer does.
I've had good luck aiming for easier targets. Android APKs, Java, and .NET binaries are usually easy to reverse engineer with AI and without any advanced tooling.
Without admitting to anything, I was surprised by how little time it took to add support for the new protocol to the library. And I didn't even need an MCP server.
I feel like I'm either too dumb to get this, or not the target audience. I don't know what a Tapo is, but it's apparently a library to interact with Tapo devices, whatever that might be. And now it can talk TPAP, whatever that is. TP-Link is mentioned, so I'll guess Routers and move on :)
Tapo devices are basically smart home stuff. You normally have to use their ecosystem to control those devices, i.e. use their apps etc. However, people do not like the fact that you're locked in those apps and want to use those devices via platforms like Homeassistant, so people figure out how to do it by tricks like this.
Author here. tapo is an unofficial Rust client library for TP-Link Tapo devices (plugs, lights, hubs, cameras), with a Python wrapper built on the same crate. It is not affiliated with TP-Link.
The short version: since late 2025, firmware updates have made Tapo devices refuse third-party clients unless you turn on a "Third-Party Compatibility" switch in the Tapo app. The switch works by bringing back the older login, KLAP. With it off, devices speak an undocumented protocol called TPAP, which logs in with SPAKE2+ (RFC 9383). The library now speaks TPAP, so the switch can stay off.
The part I found most interesting is the security difference. A recorded KLAP login can be used to test password guesses offline. With SPAKE2+ it can't, and learning the password later doesn't decrypt sessions captured earlier. So the "compatibility" switch is really a security downgrade, and TP-Link's own FAQ says enabling it "may reduce the security of your devices".
Not everything works with the switch off yet: some cameras, such as a C210 on firmware 1.5.2, still need it on.
Happy to answer questions about the protocol work or the library.
I have a handful of old TP-Link wifi switch devices, but I haven't kept up on the play by play developments. I just know at some point newer ones stopped working with that access method (and I haven't bought any since).
Is KLAP that old local-network UDP protocol with "XOR encryption" ? Or is that something else?
Does using TPAP with your library still require connecting the devices to their "cloud" (warning: surveillance!) ? Or does your library effectively restore the local-only workflow of never allowing the devices Internet access, and controlling them locally ?
That XOR protocol is a different one, and older than anything in the article. It is the original TP-Link Smart Home protocol used by the Kasa line (HS100, HS110 and similar): JSON on port 9999, obfuscated with an XOR autokey cipher, with no authentication at all.
Tapo devices never spoke it. Their original protocol was an AES passthrough over HTTP, KLAP replaced that in 2023, and TPAP is the newest. As far as I know, newer Kasa hardware and firmware moved to KLAP as well, which would explain why your access method stopped working on the newer ones. My library only covers Tapo devices; for Kasa, python-kasa is the one to look at.
On the cloud: the devices do have to be set up through the phone app with a TP-Link cloud account. Once set up, though, most functions of most devices can be used locally through the library, with neither the devices nor the library having internet access. The main catch is credentials: if you change the account password, for example, the devices need to be online for a little while to pick up the new one.
I don't even really care AI or human if its written like this. I would rather do anything else than continue reading.
It is either satire designed to grab engagement by shocking people, or the genuine anti-AI frustration of a mediocre writer who has realised they've lost their edge, because their work isn't worth paying for over AI.
The people winning literary awards will keep winning them with their manual writing, and they will keep making good money ghostwriting.
But my article, which is meant to communicate some software features in decent, readable prose, isn't worth $10k of ghostwriting. AI will have to do. So you go get, sir, and please stop gatekeeping! (Not you specifically, you get my drift.)
The Economist recently published a piece on this very subject, too, and I found it a good listen: https://www.economist.com/by-invitation/2026/09/07/this-arti...
And there's no excuse now anyway the latest Opus/Astra models are actually tolerable. Use those if you must.
Not being able to effectively proofread will make slop even sloppier.
There's no reason to be this nasty. If you don't like the writing, skip the post.
For what it's worth, the writing is clear and it gets the point across.
OP, thank you for doing the work and for sharing it.
I can bet author didn't even read his own AI generated text.... I don't have any issue with AI, but it would be nice if people spend time over creating quality stuff.
now all you need is IDA or Ghidra MCP, a binary and some vague sloppy instructions.
some more recent models even started instrumenting a running binary (when possible) to enumerate the protocol without being explicitly instructed to, which is even better.
Seriously, though, the popular Ghidra MCP is really badly architected; it's way better to rearchitect it or just script Ghidra directly. With that said, Opus 5.5 seems to have been trained on CoT from the popular Ghidra MCP. This makes it work better, but also makes it even more inefficient if you modify the MCP without changing its name and shape significantly (it will try to make tool calls using the "mainline" format, then have to retry them when they fail).
Even with Opus 5.5, IMO it's better to just ditch the MCP and let the LLMs eat with bintools and headless Ghidra; with both GLM and Opus this produces significantly more efficient results than the popular MCP. On the other hand the IDA Pro MCP is much better architected and seems to be pretty good.
I don't use Ghidra but IDA Pro MCP works extremely well for me for all manner of tasks. for example, some software likes to call home for license checking (and I wish to run it with networking denied to it). it no longer does.
>Tapo is a brand of TP-Link that offers smart home solutions, such as security cameras, plugs, bulbs, switches, and more.
and TPAP is the "TP-Link Authentication Protocol" which, i believe, is based on 802.1X (https://en.wikipedia.org/wiki/IEEE_802.1X)
The short version: since late 2025, firmware updates have made Tapo devices refuse third-party clients unless you turn on a "Third-Party Compatibility" switch in the Tapo app. The switch works by bringing back the older login, KLAP. With it off, devices speak an undocumented protocol called TPAP, which logs in with SPAKE2+ (RFC 9383). The library now speaks TPAP, so the switch can stay off.
The part I found most interesting is the security difference. A recorded KLAP login can be used to test password guesses offline. With SPAKE2+ it can't, and learning the password later doesn't decrypt sessions captured earlier. So the "compatibility" switch is really a security downgrade, and TP-Link's own FAQ says enabling it "may reduce the security of your devices".
Not everything works with the switch off yet: some cameras, such as a C210 on firmware 1.5.2, still need it on.
Happy to answer questions about the protocol work or the library.
Great work!
Is KLAP that old local-network UDP protocol with "XOR encryption" ? Or is that something else?
Does using TPAP with your library still require connecting the devices to their "cloud" (warning: surveillance!) ? Or does your library effectively restore the local-only workflow of never allowing the devices Internet access, and controlling them locally ?
Tapo devices never spoke it. Their original protocol was an AES passthrough over HTTP, KLAP replaced that in 2023, and TPAP is the newest. As far as I know, newer Kasa hardware and firmware moved to KLAP as well, which would explain why your access method stopped working on the newer ones. My library only covers Tapo devices; for Kasa, python-kasa is the one to look at.
On the cloud: the devices do have to be set up through the phone app with a TP-Link cloud account. Once set up, though, most functions of most devices can be used locally through the library, with neither the devices nor the library having internet access. The main catch is credentials: if you change the account password, for example, the devices need to be online for a little while to pick up the new one.