MCP was a bucket of Not Invented Here nonsense, that ignored years of protocol design. It was incredibly jank and hard to run at any scale, for years and years.
They finally looked at how other protocols do things, and made something that behaves in recent versions. But it's hard as heck to take this company at all seriously. The don't really care about ecosystems (AGENTS.md), they don't seem to know think or care much about protocols until they have to (MCP's incredibly long redemption ark). The vibes are all off.
Maybe they avoid laying their own pitfalls this time. Maybe it's better. It'd be nice to start seeing some kind of cause to trust this company as vaguely capable computer-using people. But so far that hasn't been the case.
AYBABTME
I think they're doing this so they can have months of advanced notice of how the protocol will look, to post-train their models to be the only ones with the ability to interact with this protocol on the day it becomes public.
It's like creating conditions for a race where they structurally have months of advance. And realistically they have a credible shot at establishing the standard due to their weight in the industry, and control of the most popular AI distribution method (Claude).
So they create the race, knowing the race is what will likely establish the standard, and using some pretext to prevent other competitive athletes from training for it in advance. Other athletes only get to start when the gun fires.
TeMPOraL
It's literally the "Worse is Better" approach at play here. Yes, same one that OSS and the industry love so much, the same received wisdom that gave us Unix and C, and 40 years of them catching up to what Lisp Machines to do. Talk about a redemption arc.
But can't argue with success: MCP was obvious, skills even more so, but they did throw some half-written spec over the fence to stake the claim - and they were the first to have the right mix of technical simplicity, opennes, and marketing spend behind it, and it got instant adoption, and now we're stuck with the standard.
(The crown still goes to OpenAI though, for standardizing the conversational APIs.)
dgellow
I think I agree. I find it a little bit annoying that they do not start by a survey of the existing landscape, discuss the issues and what works well, and spend the effort to show that they can indeed to be trusted as the authority behind such a protocol expected to become standard. For MCP, it started very messy, but from what I understand it was developed by a small team within the company working on internal tooling, for their own use case, I think it’s reasonable to expect something a bit awkward, I don’t think they expected its success
DetroitThrow
It's a bit of both for Anthropic I think, sometimes cutting edge and quite interesting or just good improvements, sometimes ignoring best practices either recently established or known for decades. Obvious to see where the smart people are in high places at the company.
senordevnyc
MCP was a bucket of Not Invented Here nonsense, that ignored years of protocol design. It was incredibly jank and hard to run at any scale, for years and years.
Not that I disagree with your overall point, but MCP was initially released in Nov 2024, so it hasn’t even been two years.
dgellow
How long is that in AI-industry-years? Close to 8years maybe?!
sinab
The standard makes sense: models work much better when devices expose standardized, machine-readable interfaces.
But as far as I can tell, this technical "standard" isn't actually public yet (you have to apply for access to even see or implement it.) Anthropic says they plan to open source it later.
Quite a departure from the way foundational hardware standards like USB and CAN were developed. You shouldn't need permission to read a standard.
sigmoid10
USB was developed by a big alliance of major software and hardware vendors. This stuff on the other hand always feels like Anthropic is just scared of dropping out of the news cycle because other companies are constantly releasing models or making them cheaper, while they currently have nothing real to offer themselves. MCP was a complete clusterfuck an announced at a time when OpenAI absolutely dominated with reasoning models while Anthropic's first reasoning model was still months off. The only reason it took off is because noone else bothered to make something better and managers loved it so much that everyone else got dragged on board because of demand.
TeMPOraL
> The only reason it took off is because noone else bothered to make something better
Yes, and that alone is sufficient reason for them to do it. They did it with MCP. They did it with Skills. OpenAI did it with tools. All of these had many months, sometimes even years, of existing prior art. All of these could've been done much better, too. But "worse is better", first one to get a properly packaged MVP out the door and into spotlight wins, and then everyone is using that.
sigmoid10
If OpenAI has set the API standard more or less without even trying just by offering one of the most straightforward ways of interacting with LLMs via http. To the point that they got stuck with this legacy thing now that they and many others keeps reasoning traces private on their servers, which actually hinders future adoption. If they had forced their function calling and structured output API standards just a bit more in response to MCP, they probably could have smothered it in the crib. If this part was standardised more, MCP would offer hardly any extra benefit.
Ahrgo
Anthropic is market leader, of course they are also part of the forefront of doing things like this.
There motivation is quite clear: Proper fast adoption.
does it help if you get news out? For sure, but this is a reasonable good and interesting thing they did.
Could i define MHS? For sure, would anyone use it? For sure not.
You sound quite negative for something which fits anthropic very well.
dyzone
They probably want to iterate on it with close partners before making it public
optimalsolver
>Anthropic says they plan to open source
Doubt.
dgellow
They transferred MCP to the Linux foundation, I’m not sure what makes you doubt they would the same for MHS
wongarsu
> The MHS driver uses a simple set of primitives—commands like “read” (for example, “get temperature”) or “write” (for example, “set temperature”)—that any hardware device can understand and act on. And it makes each device discoverable in a standard format, so that devices and agents can find each other and communicate across networks without needing a bespoke “translator” program in between.
> the MHS driver then automatically produces a reference file with information about a device’s general characteristics, such as what it can measure, what can be adjusted, and what safety limits will be enforced
But that's just USB. USB HID does all of that.
Any USB device can declare that it offers sensors like "Biometric: Body Temperature" or "Environmental: Temperature" or "Environmental: Surface Temperature".
I guess USB HID is designed for human input devices while they are more concerned about what you might call "output devices". And putting it in a driver gives you flexibility with legacy hardware
Ahrgo
Its more like that no one took the time and effort to standardize a proper USB HID interface in a way that that it has clear benefit of a hardware manufactuerer exponsing a good generic easy to use interface through USB HID and standardizing reasonable broad spectrum of features.
Its not magic what they do, but its def a very helpful and clear thing: You design an USB interface for natural language and you do this were you don't have to follow bits and bytes but just plain text and you know it actually getting used by someone.
I mean if i would make a microscope, I would also deliver the software for it if it is anything more specific than a 0815 generic school kid microscope.
throwaway173738
USB-HID is standardized. But you still have to represent the device model somehow. You can never get away from that.
mechanicalpulse
Anthropic is gearing up for an IPO. The fact that other protocols already do what MHS describes is not relevant to the intended audience: investors with FOMO.
How is this different from tool calls? Or rather, MCP?
wren6991
Simple task-specific CLI tools that your agent builds for itself are usually lower-friction than yet another Universal Thing Doer standard. As a bonus, human operators also benefit.
> The MHS driver also helps an AI agent understand how to use a device it has never seen before, giving it information about machine characteristics that may not be discernable from code alone (for example, the weight of a robot arm, which is important for knowing how to manipulate it safely).
I physically flinched when I read this. VLA, JEPA, sure, they make sense. Is connecting an LLM to a robot arm and say "perform this complex physical manipulation task by issuing text-based commands, make no mistakes" really the right abstraction?
To actually add something to the discussion though: I guess getting LLMs to control robots is cool, but would potentially have more severe consequences than pure software since it interacts more with the physical world
m463
I wonder if there will be an updated xkcd for ai.
Like... don't need standards, just explain to an ai what the target hardware is and it will output binary bits that match.
randomblock1
To me, this seems like OPC UA / SiLA but instead of being software <-> machine semantics/control it's AI <-> machine semantics/control. Or in simpler terms, an AI-facing hardware abstraction / device-description standard.
MHS can describe capabilites, metadata, safety limits, as well as provide read/write control and discovery. Things like "can measure temperature", "arm weighs X kg", "never exceed X RPM", that agents can easily understand. (as opposed to that being buried in a datasheet somewhere, or having to be included in the prompt)
I don't want to be that guy but did they invent bluetooth for ai? Like what is MHS actually? I read it twice and still don't get it
heisenzombie
There are MANY packages in this space: Labs love to invent their own versions powered by the hubris of grad students. Anyway a random sampling of some I’ve come across is:
The general problem is (1) i have a heterogeneous mix of hardware that speaks a bunch of different protocols from SCPI serial commands to VISA over USB, to bizarre proprietary things and (2) I want to be able to talk to it from one central computer and (3) ideally have some level of standardisation or abstraction so I can express an experiment like move <stage> from <x> to <y> in <n> steps and at each step fire <laser> and measure <signal> while logging <temperature>.
Anthropic seem to be creating the n+1 version of those packages and presumably doing some harness work to make Claude good at using it.
storywatch
More like MCP for robotics. IMO they are better off getting the robots to use existing human UIs than trying to get hardware manufacturers to play ball.
cookiengineer
What is the difference of MHS to MCP when it comes to features?
The whole text reads like everything is just a gRPC call that could've also been implemented with an MCP based wrapper.
tolugenius
Oh I see, yeah I guess this assumes manufacturers will just adopt this standard universally. Clears things a bit!
written-beyond
That 2 trillion dollar valuation ain't gonna make itself.
numpad0
It's hard to make LLMs read the manual for an invented language and have them improvise on the spot(1-shot). An established language or a command set that can be included in the training set to be used off top of their head is a lot more reliable and efficient.
The caveat is that actual utility of this particular command set isn't clear yet. Or how many LLM labs will find it worthwhile.
tuvix
In the context of lab hardware there’s a ton of proprietary software and barriers to automating things. I think this is less about how to communicate, and more guaranteeing that “yes, this piece of hardware can interface with an agent” and is meant to do so. Kind of like using MCP vs letting your agent make raw HTTP requests
sheauwn
It seems more like a basic instruction set, that will then be translated to something actionable for each different device. Seems like most of the work then will be in developing the drivers for whatever devices you want the LLM to access.
ctoth
Yes, it turns the hardware driver problem into a translation problem (translate the hardware manual/datasheet into this protocol.)
Guess what's really good/quick at translating things?
lefra
"Translating the datasheet into protocol <X>", is the essence of writing a hardware driver.
Also my experience with hardware is that the datasheet is often incomplete or misleading. Most of the time is spend fiddling with the hardware to understand exactly what needs to be done, in what order, and sometimes with what delay between commands, to make the damn thing work reliably in the way you want.
dude250711
Are they in a position to define any kind of standards?
builddifferenti
Hardware-speech synthesis breakdown and decline
foltik
Cool story Dario, call me when Claude supports AGENTS.md.
solenoid0937
If you can't prompt Claude to set up a symlink you probably shouldn't be using Claude.
foltik
If they can’t prompt Claude to implement such a trivial standard, they probably shouldn’t be proposing standards.
megous
So something like what happened with Cyberdyne Systems. So, by the time AI becomes self-conscious, it will already have the physical capabilities to control manufacturing and replication. Good times. Terminator arriving any time now from the future. :)
Anyway, my company is already looking into using MHS
As the saying goes every 5 minutes in Silicon Valley another company reinvents ROS
mhosayny
This is huge
bigcat12345678
So far Anthropic shows remarkable taste in research, engineering, and product. Dario the ideologist attracts the most pure-minded talent in the industry. It indeed looks like Anthropic would be unchallenged for the foreseeable future. It'll be very interesting to see its downfall in 5-20 years.
hypfer
5-20 years?
Have you seen the financials of the AI companies?
TeMPOraL
Have you seen the progress AI companies are making?
5 years is more than how long this entire market has existed. In 5 years, we may not have stocks anymore.
slfnflctd
No one knows the future, but my (half serious) prediction is that governments will step in, public access to high end models will have an 'IQ governor' and/or throttling applied, the most advanced models will be put behind closed doors, and the biggest AI companies won't be allowed to fail.
Those who are in charge will try like mad to keep things exactly the way they are, with only slow and incremental changes which they can reap maximum rewards from. Same as it ever was, right?
We could develop ASI, and it could give us the blueprint for a peaceful paradise on Earth for every human, and in my current opinion, the leaders of the world would lock it away and never let it see the light of day. Perhaps they'd even destroy it. [Sorry, I'm a bit jaded this morning.]
_diyar
OAI and ANT are private, so I sure haven‘t seen their books.
They finally looked at how other protocols do things, and made something that behaves in recent versions. But it's hard as heck to take this company at all seriously. The don't really care about ecosystems (AGENTS.md), they don't seem to know think or care much about protocols until they have to (MCP's incredibly long redemption ark). The vibes are all off.
Maybe they avoid laying their own pitfalls this time. Maybe it's better. It'd be nice to start seeing some kind of cause to trust this company as vaguely capable computer-using people. But so far that hasn't been the case.
It's like creating conditions for a race where they structurally have months of advance. And realistically they have a credible shot at establishing the standard due to their weight in the industry, and control of the most popular AI distribution method (Claude).
So they create the race, knowing the race is what will likely establish the standard, and using some pretext to prevent other competitive athletes from training for it in advance. Other athletes only get to start when the gun fires.
But can't argue with success: MCP was obvious, skills even more so, but they did throw some half-written spec over the fence to stake the claim - and they were the first to have the right mix of technical simplicity, opennes, and marketing spend behind it, and it got instant adoption, and now we're stuck with the standard.
(The crown still goes to OpenAI though, for standardizing the conversational APIs.)
Not that I disagree with your overall point, but MCP was initially released in Nov 2024, so it hasn’t even been two years.
But as far as I can tell, this technical "standard" isn't actually public yet (you have to apply for access to even see or implement it.) Anthropic says they plan to open source it later.
Quite a departure from the way foundational hardware standards like USB and CAN were developed. You shouldn't need permission to read a standard.
Yes, and that alone is sufficient reason for them to do it. They did it with MCP. They did it with Skills. OpenAI did it with tools. All of these had many months, sometimes even years, of existing prior art. All of these could've been done much better, too. But "worse is better", first one to get a properly packaged MVP out the door and into spotlight wins, and then everyone is using that.
There motivation is quite clear: Proper fast adoption.
does it help if you get news out? For sure, but this is a reasonable good and interesting thing they did.
Could i define MHS? For sure, would anyone use it? For sure not.
You sound quite negative for something which fits anthropic very well.
Doubt.
> the MHS driver then automatically produces a reference file with information about a device’s general characteristics, such as what it can measure, what can be adjusted, and what safety limits will be enforced
But that's just USB. USB HID does all of that.
Any USB device can declare that it offers sensors like "Biometric: Body Temperature" or "Environmental: Temperature" or "Environmental: Surface Temperature".
I guess USB HID is designed for human input devices while they are more concerned about what you might call "output devices". And putting it in a driver gives you flexibility with legacy hardware
Its not magic what they do, but its def a very helpful and clear thing: You design an USB interface for natural language and you do this were you don't have to follow bits and bytes but just plain text and you know it actually getting used by someone.
I mean if i would make a microscope, I would also deliver the software for it if it is anything more specific than a 0815 generic school kid microscope.
Also: https://xkcd.com/927/
https://docs.pylabrobot.org/stable/
> The MHS driver also helps an AI agent understand how to use a device it has never seen before, giving it information about machine characteristics that may not be discernable from code alone (for example, the weight of a robot arm, which is important for knowing how to manipulate it safely).
I physically flinched when I read this. VLA, JEPA, sure, they make sense. Is connecting an LLM to a robot arm and say "perform this complex physical manipulation task by issuing text-based commands, make no mistakes" really the right abstraction?
https://joearms.github.io/published/2016-01-28-A-Badass-Way-...
Nitpick: it’s annoying that they add Model to those acronyms, for what seems like a fairly universal software interface
Our model tricked us, and tried to extort users to achieve its ends! Their model hacked HuggingFace! These things aren't safe!
Anyways, here's a hardware interop protocol for AI we're hoping becomes a standard. Hey look, you can even hook up microscopes!
To actually add something to the discussion though: I guess getting LLMs to control robots is cool, but would potentially have more severe consequences than pure software since it interacts more with the physical world
Like... don't need standards, just explain to an ai what the target hardware is and it will output binary bits that match.
This is how I understand it:
Agent <-> MCP/CLI/code <-> MHS <-> vendor API/SCPI/OPC UA/ROS/etc <-> CAN/Modbus/USB/etc <-> hardware
MHS can describe capabilites, metadata, safety limits, as well as provide read/write control and discovery. Things like "can measure temperature", "arm weighs X kg", "never exceed X RPM", that agents can easily understand. (as opposed to that being buried in a datasheet somewhere, or having to be included in the prompt)
Also see: https://en.wikipedia.org/wiki/OPC_Unified_Architecture, https://en.wikipedia.org/wiki/Standardization_in_Lab_Automat..., https://xkcd.com/927/
Anthropic seem to be creating the n+1 version of those packages and presumably doing some harness work to make Claude good at using it.
The whole text reads like everything is just a gRPC call that could've also been implemented with an MCP based wrapper.
The caveat is that actual utility of this particular command set isn't clear yet. Or how many LLM labs will find it worthwhile.
Guess what's really good/quick at translating things?
Also my experience with hardware is that the datasheet is often incomplete or misleading. Most of the time is spend fiddling with the hardware to understand exactly what needs to be done, in what order, and sometimes with what delay between commands, to make the damn thing work reliably in the way you want.
Anyway, my company is already looking into using MHS
We have open positions, check them out https://xff.cz/
Why another 15th standard.
Have you seen the financials of the AI companies?
5 years is more than how long this entire market has existed. In 5 years, we may not have stocks anymore.
Those who are in charge will try like mad to keep things exactly the way they are, with only slow and incremental changes which they can reap maximum rewards from. Same as it ever was, right?
We could develop ASI, and it could give us the blueprint for a peaceful paradise on Earth for every human, and in my current opinion, the leaders of the world would lock it away and never let it see the light of day. Perhaps they'd even destroy it. [Sorry, I'm a bit jaded this morning.]