MCP, David Cramer from Sentry, A2A, Dave Yen from Orange Collective

what's up everybody we're coming at you live i'm Shane this is This is Obby where are we coming at you from coming at everyone from we're here with the Admirals of AI and we're at the Palace of the Dog right here sponsored by Principles of Building AI agents not sponsored by Celsius but maybe one day how's it going dude dude it's going

good we had some sick demos today in our team meeting yeah we should talk about those yeah well what's what was the highlight for you what was the coolest thing you saw there's like a lot of um there's there was I think A to We should probably talk about that in more detail as well um but MCP hosting that

we're doing yeah we're gonna have MCP hosting maybe we should ask Daniel to come in at some point in the stream and kind of talk about that yeah we could we could bring Daniel in I think there's you know some some other people that have some some hot takes around MCP as well yeah there's a lot of hot takes in there i know hey have you ever seen like

you know I want to get into the the show but I just been thinking about what I'm going to do this weekend like have you seen that new Marvel movie no the Thunderbolts i honestly I've kind of fallen off the Marvel universe a little bit but I have not seen Thunderbolts no there's this like new um hero in it called the the century you know I don't

even know about it the century okay yeah but it made me think of one of my friends i think it's a friend of ours right david yeah i've never really talked to David but I know you have yeah i think maybe we should do phone a friend phone a friend let's phone him in oh there he is that was quick hey David so cringe yes yeah we we

lean into the cringe here we're we're we're not above a little cringe uh yeah let's Yeah it's good to good to see you David yeah bring you on and maybe uh just chat about some MCP stuff i saw you had some blog posts come out the last couple days on on MCP so I figured we can spend a little time talking about that and see where it goes

from there yeah I've been full uh Devril which is not my job to be clear i run a company sort of or I help run a company but um it's been fun may maybe yeah maybe just quick background maybe that would be helpful a lot of people probably know who you are but maybe not everybody uh if you don't know me I'm

the founder of Sentry hopefully you know what Sentry is if not you know it's not you know you just go check yourself it's it's not a good thing but uh uh yeah I run uh I don't actually know what I do i do a lot of random stuff but um last couple weeks I've been building the MCP server for us which is not usually what

I do to be fair i haven't written code for the company in 8 years uh until this so uh but it's been fun i don't know um what do you think the like I kind of want to start where like why do you think this thing took off so so quickly is it is it because of companies is it because of the model company backing it or some other phenomenon no I think it's

everybody wants to like it's like FOMO everybody wants to be like pick me pick me pick me I'm here too they want to be relevant you know and I do think it's like MCP is not a bad idea mind you it's you know my my latest rant was it's just not that good right now um it's a good architecture and it makes a lot of sense there's still tons of unanswered

problems but it is literally if you look at So when I was writing this post about it the other day I Googled MCP and I bet like I'm going to pull it up on my own screen uh I bet if you Google MCP right now you're going to get a sponsored listing oh I didn't get one right now uh I got one for um what was the company it was this random Java company uh uh Neo4j

it's like some old school graph database but there's a sponsored listing for the the term MCP and I'm like you know what is going on here and I don't know so you can clearly see it as very much like latching on to whatever's happening right now which is to be fair it's like good from a go to market point of view I guess but a lot of the stuff doesn't make sense like you don't need to stick

all these random things in a chatbot it's just like that's not how you consume it you know for some people it does though it makes a lot like Century makes a lot of sense so yeah so we were talking to some YC founders yesterday that they were not they were MCP adjacent before MCP so then unfortunately like you know the the hype

kind of like dissolves other protocols now they're like trying to maybe get into the MCP space they're talking to big companies they talked to um what was it asana Yeah let's just put like uh they they talk to companies like that who have MCP servers with zero usage because for like Asana or linear I think the UI has already been optimized for so many years for like a use like a user right so the MCP is not

there at least at least not yet I think the promise is there right that that's what everyone buys into this hype that this promise could exist but it doesn't exist not today yeah I mean I I think It it's not like a normal product is a thing like you're building for what we describe as like an unknown unknown or

like a known unknown one of the two but like you know there's probably value in what your company provides or at least you want to believe there is like one of the two it doesn't actually matter um you don't know what that value is in the context of these agents right you kind of want to enable people to develop but it's like APIs it's like you kind of

have an idea what the value is or at least you should and you build an API because that value is there like if you're like I'm trying to pick an obscure company but like there's a bunch of companies where like the API is irrelevant like you don't need a public API to consume it that's not the value it's offering right yep but nobody took that exercise with MCP they're like "Oh

hype let's go we can't miss this opportunity." But it's the same thing it's an API and so like ours is like "Oh Sentry you probably want to do some stuff with like the air data and all this other stuff that's in Sentry inside of something like cursor totally makes sense so of course we want to expose that there if if we can find a way to do it well I just think there's a and and

that's fine you know we at least have a thesis behind it and I think it's okay if you don't quite know how it's going to be used but I just think too many people are like "We have no clues we'll just here's our API good luck have fun." You know it was just that's not good enough where we found it very helpful so

far um is in documentation and helping people write code in IDE exactly and so like for example we have Monster MCP doc server and you know we put all of our docs in there uh we pre-process them in there and people can use that to essentially have an assistant while running with MRA um outside of that use case I've not used an MCP server for

anything other than like things like that that help me in IDEs right we we've played around with them with like Composio and all these different ones that help you uh but like yeah it's just kind of like a mess I guess yeah one one core area so we have we have a bunch of like century is a it's a big thing right

like we have a lot going on um and we have one use internally that uses LLM that does this um root cause analysis it basically summarizes all of this data sort of associated with one entity like a big graph traversal thing um and that uses a bunch of tools right and then it tries to generate a patch which uses a bunch of tools and if you think about

those tools it's like okay summarize a bunch of data it's probably a bunch of different tools that are doing summarization right one might be summarize error summarize trace all all these things like that might be useful to another agent or the codegen stuff might be useful to other agents and so one cool thing about MCP if you ignore the hype is internal usage is actually

very high value too because you can reuse this technology it's this it's like microservices right for better or worse um and I don't think people again I think people are like latching on to hype versus like thinking critically about like actually what is a useful thing I can do with this because actually there's a lot of uses for it i think it's super super valuable it's

just really rough and broken for most consumers right now today so yeah do you mind showing us uh showing off the Sentry MCP server not to put you on the spot or anything be cool to uh Yeah i have no idea how this uh thing we're on works but I can uh give it a shot yeah yeah i mean welcome to the Welcome to the club there uh there should be a stream button though or share screen

button let's see oh yeah yeah I see it okay um oh this is going to be Oh actually we can allow I'm like is this one of those things where I got to reboot the browser to give it permission yeah hopefully not uh hopefully not but we are pretty good at stalling if we need to go that route yeah we could stall entire screen okay I think we're

good all right uh we're going to do the um Yeah let's do it live the recursive screen also we have 72 people in the in here and a bunch of people in the chat we'll get to your comments in a bit yeah if drop drop comments in the chat along the way we're with David Kramer from Century he's showing off he's going to show us the Century uh MCP i guess I don't

actually need to set this up i just need something that's functioning and fingers crossed cursor does not break here let me swap this um you're So you're a cursor guy yeah I'm a cursor guy strong believer it is the best in the industry right now ob's a wind surfer so I tried it i've tried it a couple times

and it's got like some nice UI polish i think I think cursor love cursor and they know this but I think cursor has the worst UI um but I think it's got the best tech like in in my case and this is not scientific but I've used all of them and mostly using all of them to test the different things and every single time

I'm back at cursor cuz I'm like everything else is just not working as well like the type ahead is so good in cursor doesn't mean it's perfect there's lots of problems too um the agent mode actually just seems to work a lot better for me than it did in like wind surf and VS Code there's just like all these

little nuances that happen and maybe that's just like specific to what I do but generally speaking I'm like I don't know it seems uh seems to be cursor oh yeah I guess I don't need to bring up that screen to talk i'm not a streamer sorry u but I do need this screen okay you can you can share if you want we're just having fun anyways yeah yeah

um so where do I go um okay so we got this MCP server if you don't know what it is you know it's it's just a wrapper around your API and so it's a wrapper that takes something you're doing could be a public API could be a database something um converts it to in our case we're using markdown um converts it to markdown and hopefully you can do

something with that so a really good example um let's see if I can avoid PII this is always like the risk i think we're good though um if not I'm going to ask for forgiveness so let's take um which one least likely has PI let's go with this one if you want me to like you know we're probably good it's it's

it's good somebody will be like oh an IP address or something but it's actually not a big deal um okay so I'm I'm cheating a little bit don't don't worry about what's happening right now this is one of our agents that you're seeing on the screen this is our um it's going to be renamed but this is our issue fix thing what this is doing is root causing a

problem it may or may not do something useful here i just need an example to show you why MCPs are interesting um okay so we want this actually we can just take the URL let's get rid of all that crud um so let's say this is this is in this is the MCP service um set up to debug the MCP service just so we're clear on what's going on

um and we are in cursor and we've got this configured here um if you haven't configured these they're kind of they're a little little brittle um but when you got a green light it's working you won't always get a green light it's complicated all these work a little bit differently it will probably get better in a few weeks uh across the industry

but we'll see um okay so we did this thing uh I'm trying to fix a bug I found in Sentry uh can you root cause it and help me uh find a solution fingers crossed live demo i don't know actually know if this is going to do what I wanted to do cuz we're sort of banking on the agent prompt signaling correctly um but but this prompt you can this thought that I just put in into

chat is actually how you would think about or how I think about it right and I'm pretty sure this is not whatsoever doing what I wanted it to do so let's try again use centuries issue fix behavior this is fine i guess like we should not expect this is not what I wanted to do either i didn't want it to begin so here's the problem um MCPS are

super limited right now and it takes sometimes a couple minutes to do this thing in Sentry this uh thing that our agent does and so I try to signal to the agent that hey don't kick off an issue fix right away check to see if one's in progress first yeah because then I can cheat and I can start it in the background or something like that but yes pull until it is ready let's see

what happens here usually it gives up after a few tries so maybe it'll actually work this time are you calling an LLM within the MCP tool effectively yeah but it's it's exposed in this like complicated abstraction so I'm calling an API on Sentry which is running an agent behind the scenes yeah um it's

like I'll continue pulling but it's actually not continuing pulling it just thinks it is and so try again and wait a few seconds in between i don't think for example though it doesn't know how to wait so you'd have to expose like a weight tool cursor would otherwise it can't actually do any of this right okay cool it finished so um oh it's not done yet it's not done

but it has partial information which is what it's giving here and so it's like okay the root cause of is this blah blah blah um I I didn't actually look at this error so I have no idea what it is um unfortunately it's it's like oh I need to keep pulling pull again but you can understand why this is not great today though just as an example of a really good use case for

this and it's still bad is it done yet though that's the real question yeah wouldn't it be better to start this like instead of like uh prompting it like wouldn't it be or would it be better to let's say click a button that like fires off like a lot of this rooaus yeah like like people are too wrapped up in chat bots and stuff thinking that's the solution um yeah I mean I think yeah I mean I'm not the one

who came up with this clearly but I've heard you know chat bots are like the the CLI of of you know AI so it's it's fine but it doesn't it's in most cases it doesn't seem like it's the best use for Yeah because I mean like you look at this autofix thing in Sentry this is an agent it's not a chatbot like there's some text inputs in here but like down

here we've got um where is it hang on i need to move this little screen uh I think it's in here somewhere so diagnosing this is this is wrong this is where LLM's mess up by the way is like fail to parse token response does not mean the user is missing i don't know where it came up with this idea usually it's pretty good

but sometimes there's just errors that make no sense um but yeah so we've got like an actual UI that like tries to solve so code it up will just give me a solution um again it's like you got to think about how to use these user interfaces but the concept of like I want this context inside of cursor is the important one yeah how we get it

there I don't know the only way we're getting anything anywhere right now is MCP that's exciting for folks so then like let's say you get the context in there is the next thing the uh fix it and then they'll write the code from the error report or from the root cause or something yeah yeah and you'll see we actually came up with like I mean it's

the same thing and so but that the idea is like I can create a PR out of this right yeah but it's it's probably wrong because it's it's an LLM right i'm going to have to change it but why why would I create a PR and I actually don't even know what checkout locally does um but really I just want it in my

editor right now i want to be like give me the potential changes you're suggesting so I can review and tweak it etc etc etc um cuz it is like totally wrong right now but um but just contrived example anyway so I think that's an interesting concept obviously there's other things you could do but it's silly because like here's an example of how every single MCP works right now and this is not real value ad

it's like show me my most recent issues in century remote MCP it's like why is this useful to me like I got a chatbot to do this when I can go click a UI that's way more consumable it's not useful and this is every MCP that's been exposed recently right and so that Yeah that's the problem i don't know again

there's lots of useful use cases if you think of as as tools and reusable tools like um maybe somebody is so here's a real real good example if you're like a a Devon or Vzero or pick one of these code generators and maybe you had a mode that's like I can deploy this to production and then you could monitor

Sentry look for errors use the MCP to call our agent or something so you could like plug in these tools from Sentry to augment a bigger workflow but you still have to design a system you don't just plug in abstract tools and pretend they're useful you know so yeah we share that same belief like uh a lot of these agentic things are deterministic with

like indeterminism within the steps which is reasoning based on stimuli right but doesn't necessarily mean you ask the agent to both come up with the plan do the plan and then expect it to have 100% accuracy of the plan um especially because it doesn't know your business doesn't know anything about the history of your company what all the

work that it took to get here which I think there's like that's like a plus for like the human part of it you know i mean yeah the context is doesn't matter how big the context is it's usually still limited right on what it can understand so yeah 100% and sure maybe you could feed it all the context in the world but are you going to pay you know

$10,000 for this bug whatever this is it's probably not even a real bug no you're not like it doesn't make any sense so I I could totally see though as you you have agents that can you know not just run when you tell it to like you know on your input but imagine you have alerts that come into Sentry and you can wire up agents to do the first

triage of that and maybe propose a and things like that could become really useful it's almost like they were a Sentry you know what I mean yeah yeah so this is what I think is is wacky i heard you all talking about uh agent to agent and stuff I think right before joining but um like agents are a good idea like

that's a good construct tools are a good idea MCP is a great way to make tools sort of composable reusable right but agents are really the value prop and everybody's like over rotate on MCP because it's like oh we can integrate our just like here's our API we're we're in the the conversation now but you're not and there's a lot of use cases for agents and I think MCP is going to be

actually way more valuable in the in the sort of people are latching on to right now when we can expose agents through it so like like you saw what I was doing here and it was bad it's like I had to tell it to check again but really it just communicate with an agent it should be able to stream that back like any other thing that's happening in here just the the protocol doesn't support

that right now and so I think once that happens you'll effectively unlock agent to agent via MCP which just means it's super pluggable and it's still high value at that point and I actually think that opens the door to a lot more free form uses because I can just expose a sentry agent that on our servers I can

parse your query that so like effectively agent the agent in cursor sends me and then I can do something much more useful with it than these abstract tools that we're just exposing to you so yeah totally what what's your take on A toa like if you're you know if Okay so here's the thing like we are deep into AAA right now more so because we have to be um and it's the same kind

of growing pains are going to happen just as MCP is going through but what's your take so far or if at all I don't know if Away is like is this like you can't make fetch a thing or whatever um when I looked at when I saw the announcement which was conveniently right after MCP was starting to get popular um and I looked at all the logos

in the like look at all these enterprise people supporting ADA it's like a bunch of irrelevant companies like these are not the companies that are leading the way in anything these are the companies trying to latch on to things that are pure enterprise have not been super relevant for many years and I'm like that's signal to me and I look for any signal I can get in anything and so

right now I don't know we'll see what happens but I don't have any confidence in that so yeah I went to a um I went to that RSA conference in SF last week which I had no business being there by the way no freaking business and I didn't even have a ticket so actually I shouldn't have said that out loud but um I went there and I was in this dis agent

security discussion and the this I think it was OASP is the the group they were doing like these are the three layers of AI agents that you you and your teams need to know about one single agent architecture this is what we recommend which is like all right isn't that very limiting okay sure multi- aent

supervisor architecture it's like okay I can get with that then a third one peer-to-peer architecture a toa the future okay I can also get with that um or I can try to understand it but the thing that the difference is with MCP agent to agent is via tool calling and with a toa it's via this task job system right agents send tasks to each other

versus tool calls you know if you just look at it like that those are just different transports of doing the same yeah it is one one is also much more complex to implement by the way like this was my point because like uh MCP is also you know I like the guys building it i'm a fan but like it's also needlessly complicated like I cannot like the way it works is I send in a request um it goes through one route and

then the response comes through another like another system at the end of the It's like a it's it's a multi- channelannel thing so it is like fire task over here and then receive a response over here and that is so complicated to reason about when we have streaming we have web a like we have HTTP we have all these protocols that exist and my biggest complaint is I'm like you really don't need to reinvent the core protocol for the 99% use case

to support some 1% outlier over here and yeah sure maybe not all maybe a lot of stuff does have to fire off and then check back later that could still be a stream so there's some things there but I I just think it's like needlessly complicated and I I wish it was not and this is why for example out of the box it's not that easy to just support streaming responses for agent to agent

in FCP and I I wish they would do something about it and just dial it back but I know there's already been discussions and they're like "No there's use cases for this." And this is how all these design by committee things go these big academic concepts and you know it's not that it's invalid it's just it's complex and not everybody needs that complexity and so hopefully the abstra abstractions make it easier but

still it's I I don't know we'll we'll see what happens so like David we've been all doing this for a long time i also remember when you started Sentry when I was uh just I I was more of a younger engineer back then i'm not that young though or whatever but um and I remember like the community we were in was like this JavaScript community and I

feel like we've had like these types of technological movements not like at this scale of course but it's always been that design by community let's go through memory lane real quick right we have like early JavaScript frameworks Backbone Fights and Angular and React meteorjs who remembers Meteor freaking

JS of it time never used it head of its time and then all those um Okay React now we're doing React Nex.js Gatsby of course like all this and do you think like now with the TypeScript getting into AI and stuff do you think we're going to get into one a lot of more drama again i don't know here's here's my kind of point of view on everything first off

anything good usually comes from one or two people like their opinion it's not designed by committee right it's like it's like no we this is our view of the world that's okay if you don't agree with it and the best things come with our view of the world and that happens to be a really good view of the world that like changes a lot of things right and so I generally am anti- any version

of design by committee so that that's one i think two one thing we're seeing right now already ls are not intelligence it's just it's just pattern matching you know sure it's complicated pattern matching but that's all it is it should not be confused with any version of intelligence and an analogy you can apply from humans LLM is if it's hard

for a human to reason about it's also hard for this thing that's unintelligent to reason about because we pattern match it patterns matches the delta is we have a ton of context and sure in the the the vast infinity of compute it can have more context but that's just not the reality we live in and so when you look at things even even something modern right now like Nex.js JS it can be quite

hard to reason about what's going on in there typescript errors quite hard to reason about what's going on in there to the point where I can't solve them you think this machine is going to solve it like I'll spend hours on the same problem like and it's like we shouldn't be surprised that it's hard for the machines to solve some of these problems i guess is my point and so I do think if we want this technology to succeed it

might push the conversation of simplicity again which doesn't mean under the hood things have to be simple but the exposed experiences to say the the framework or the programming languages have to actually be pretty simple um I don't know what that looks like honestly but I'm trying to rewrite uh I have a side product right now i'm trying to rewrite it to be open AI open

API native um which then will hopefully get me to open AI native which I'll talk about um but I'm trying to rewrite it i'm trying to focus on DX like the ergonomics of code and I'm trying to like just remove every piece of boiler plate mhm the less code you have to write the simpler it is the less variable it is thus the more likely I

can code gen everything I want and not think too hard about things and so I'm spending a lot of time like how do I get this to like the the bare minimum of simplistic interfaces for like the 99% again you can have an outlier case where complexity enters the mix and maybe the human gets involved but most of the time I'm like copy paste this thing change

this behavior add a test that's copy pasting this other test it's like that is most of the crud we do right yeah and so this is kind of my litmus test of like well if I spend a lot of time to intentionally make this really really really simple like that you know a brand new contributor could figure out what's going on just by looking at it pretty

sure that any model will be able to codegen really effectively at it and again this is has nothing to do with LM how we should design things in the first place we just don't right like and so I don't know it's like hopefully this is a forcing function again because the efficiency comes from simplicity and I

don't know we'll see we'll see awesome yeah i mean you you definitely see this play out if you use any of the Yeah any of the other codegen tools like the bolt news the lovables like they use Vit over Nex.js for that reason right get farther before you hit that inevitable wall which you almost always hit when you're

you know using those tools at least I I have some people can get maybe farther than I have but it gets you really far pretty fast but it's because you start with a simpler architecture and you start and it's easier to understand the entire codebase then right and so I think this is also why it's like so effective from like zero like starting a

new project because it's simple right and I also think like Vzero is pretty effective for next but like they spend a lot of time building like basically this like the strategy is like summarized context over and over with like rags and so they spend a lot of time building these tools and these contexts that allow it to be more effective at next

but these are generalized models and so if you want it to be genuinely effective at technology things need to kind of look similar they need to look simple you know they need to not be innovative and to be fair technology doesn't change like MCP is not a new idea like it's an RPC protocol like it doesn't have to be as complex it is right like so it didn't have to reinvent a new thing the exposed

interface looks like RPC for the most part so it's easy to reason about fortunately um but yeah so I I I really hope we see more of this simplicity because if you can design simple tech hopefully you'll see like literally financial advantages to doing that thus it'll motivate the right people to make those decisions yeah dang that's awesome

thanks for coming on we have I have one more question for you before we let you go and we're going to do some boring programming after this this was dope there's this topic that um in AI right now called skill acquisition and I'll just like I'll put it in like I want to put it in the mpm J like in the JavaScript context for the viewers right so right now like as a as an if an agent what if

you could have an agent that has no tools the only tool it has is tool discovery where its job is to understand what tool to then install use authenticate maybe I have to do human in the loop I have to authenticate with you or authorize something then I execute this uh tool that I've acquired and then

I forget about it or I I actually save it like I I now I know how to do this what do you think how how crazy do you think that sound or is that a future that you would like to see i think it's a good experiment somebody should try it see what happens like here's my analogy is like we can scrape the web for docs

and they'll be out of date or you could have something that exposes the right version of the right docs to the agent all the time that's not quite what you're talking about but it's the same construct right of like I need the right information to make the decision here and I need to be able to discover that information i need a way to discover that information i think in the vastness

of the internet the idea that you have a registry for every website of all time is silly and so I don't know about that specifically but it does feel like there is something some variance of this cuz you can't again you can't just have everything in memory all the time it just doesn't work that way um I don't know what it looks like and I and I honestly think the risk the the main

thing I always tell people especially right now in the space I don't know anything about this by the way i'm just winging it and using all of my educated guesses from all of my like you know work experience in life right um but we can clearly see things change quite rapidly even if you go back like one year everybody's like "Oh we need these

DAGs of pipelines of rag systems and whatnot." And rag's still valuable i don't need a pipeline anymore i have an agent flow it's fine i it's non-deterministic already i'm not trying to make a pipeline anymore um I'm trying to like input output is the result correct right and so even that is a massive shift if you built a lot of

infrastructure and technology around one's concept and so I think it's the advice I always give to people is like be very cautious of how much you invest into something because it could change tomorrow mcp could be gone in 6 months right like who knows um but you you have to try you have to experiment and see what's going to work see what's not going to work models are going to change maybe what works now will work better

later maybe what doesn't work now will work better later i do think the concept of like you need to be be able to have a clear way to find certain things that you don't necessarily know about in advance is really really important and the way we do it today is not great like rules files or like the docs context thing that some of the systems do it's like sure it's a hack that's fine memory

memory is a hack that's fine um I do like memory it's a cool concept but um yeah so I I don't know it'd be interesting but I do think there's too many tools so it's really hard to say like would it actually function in reality tool selection is a hard job if You gave it a thousand tools to pick from yeah so be a hard job for us right yeah hard job yeah yeah exactly it's no

different if you if I gave you five tools versus you know a thousand it's going to be a lot easier and what if they're all named Fi MCP file system all of them uh someone mentioned in the chat toolbox server from Smithery i know yeah that Yeah Henry's working on that yeah so Henry over there that that's pretty cool

i I haven't used it but yeah um yeah yeah David thank you so much for being here we would love to have you on again we're just figuring this out too so we don't know what we're doing yeah it's it's fun i I like the There's a lot of hype right but I like like the Let's be authentic let's be honest we don't have

it figured out it's cool you know so it's fun uh but yeah thanks for having me on guys yeah see you leaving us right now the sentry all right did you guys like that joke to bring him on though i thought that was fire yeah I thought that was No he No he came in immediately as I told you before that thing's cringe he was cringe dude i loved it i love Hey

we're not above being cringe here and having fun you know we're just We're just two guys figuring out like all of you all probably are too yeah we don't know what the hell we're doing but yeah we I mean we still have some time so maybe we should uh Yeah we got about half bring So we have our buddy Ward um

he sent a video because we phoned a friend but he actually sent a video i have a video to play yeah can you download it already just download it from Slack all right yeah how do I do that you know we'll figure it out this is where we figure stuff out demo oh I can add a video there we go it's technology okay so it's uploading yeah and so this video give us give us the

quick preview and then we'll watch the video yeah so we are looking into A to A many of our users and many of you all wanted to see ADA working we are we built we follow the spec we um yeah we have it working essentially uh we have plans to make it better after reading the ADA spec uh and I'll get into ADA or Ward will get into ADA in

the video but after reading the spec there's just a lot to be desired it's kind of the early days of MCP um but we kind of see where it could be useful so this is kind of our experiment maybe also like David said this is us just checking off the box as well right because we have to be in the conversation yeah yeah ultimately you

know having having a framework you have people you have customers that are asking about new things that are coming out and it's actually it's hard to keep up with all this stuff right so our goal is to try to stay on the leading edge as much as we can so we can catch you all up you know if you can't follow along with everything we certainly miss stuff

um yeah live stream on LinkedIn yeah actually yeah it's all over we're on X LinkedIn YouTube oh and Huie's actually coming on the live stream yeah he left us a comment he's coming on the live stream next week if any of you all are working on something cool and you want to show show it off you know find us you

can find us Discord or X or somewhere and we're happy to bring people on show cool things teach people well that's what we're doing trying to keep people uh up to date on things this video is uh processing so it's it'll be done here in a second but ADA is is interesting right the idea of being able to communicate with agents you know from one agent to the next agent but I do think it is it

is still early days and I don't have a lot of confidence if it's going to take off or not i I don't you know I'm kind of in the same boat as David but it is I think important for us to understand it so we have you know support for it and we can then you know as things change we can kind of evolve with it

almost there how many people we had 151 hello yes 150 people watching right now um yeah we're going to do this uh quite a bit uh yeah for those of you that are just joining us I'm Shane this is Obby we're from MRA and we had David from Sentry on earlier showing us some of the Sentry MCP uh stuff that he's been working on and now we're going to talk a little bit about AD to A and see a video from Ward

from our team uh just showing a demo this is more like the actual implementation in Ma it's not finished yet so you're seeing kind of a preview of what uh what will be here hello everyone another Friday i'm going to show you the A2A protocol in Masra we now have full support for A2A and I'm just going to show you quickly how it

works so I have um a project with two separate master projects because A2A is a protocol to talk agent to agent um from Google so I want to show you how to connect two agents together so I have one weather agent which is just our simple example like hey can you tell me what weather it is so if you just do create master you get it so we have like a an agent your

helpful assistant that provides accurate weather information like so and then we have the tool who actually does the weather response as you can see here so nothing special then um I have other agent so just basic example as well just have a like my agent my agent is like there's nothing there just a helpful assistant could have done the same thing like I'm a weather assistant blah blah

blah blah blah but um I don't think it was necessary for the demo and have like a weather tool if we look at the weather tool itself so it's the same thing we're going to create a tool get the weather in a specific location we get the location and we're going to return a forecast which is just a string um I have a master client

basically it's running on localhost 411 i'm going to get the A2A client so by doing this we're getting the weather agent client and then I can do like send message there's also weather agent like get cart get task if you have like running processes happening in the background um we can do like streaming

as well with send so we have the whole protocol set up then I'm going to do the send message i get the run ID from our tool call and if it doesn't exist I can do a random U ID as well um session ID is a resource ID could be like um a user ID as well whatever you want to pass through you can it's not also not mandatory so you can um remove it as well and then we're

going to do a message so the user and we're just going to do what's the weather in location it's text and then whenever it gets back we're going to get the task and um get the result out of it this is a bit messy but this is how the protocol describes it you should mostly do like um some validation on it but for demo

purposes this is all I'm going to do um today so then I basically have two MRA um instances running so the weather agent as you can see and then here um the other example just the my agent one so if I do different port it's now running then I can go to uh my browser so first I want to show the card so in 808 there is something like agent cards and it's like on the well-known

agent name and then the agent JSON which then gives you the name of the um of the agent the description where to call it from um who the provider is um version blah blah blah all the skills so you can see the weather tool get current weather for location so if I then go to the playground as you can see I'm on 4112 so not 41 411 so this is the

my agent and if I now say like what's the weather in Paris it will call the weather tool and then hopefully give me yeah like see you can current weather in Paris blah blah blah and you can see that I got like the tool got called and then with the result and that basically did the Boom cool so there's a little sneak preview again still

uh early but you can get a sneak preview and we do have another uh someone else uh coming in here what's up Daniel how's it going and and that's not all it's a party now hello but as as someone pointed out uh ear or yesterday this is now an expensive meeting we we now we now are having a very expensive meeting but uh yeah we have some some stuff in the chat uh yeah oh what what glass speed i

guess I can do that you know you can do that yeah so yeah Glass Feed has you know TS port of Google yeah adk so that's cool that's awesome production value is through the roof now and Casey what's up dude casey what up dude hey Mr aura himself is here oh yeah all right mr aura himself all right um it's good to Yeah it's good to see you guys what

oh the music the music stopped you know when once we we we when we're vibing long enough the music decides it's ready to you know be done so we'll change it up a little bit what are you guys up to i'm I'm not sure you know what was Obby you invited some guests what's the plan dude phone a friend again we've been phoning friends all day phone a friend Friday yeah um Daniel's been working on

some exciting stuff if you guys were on the live stream yesterday we ran into some bugs that I think are maybe not bugs anymore um so Daniel take it away show the show show the crew what you got going on 172 of us here yeah you can uh you can do a lot in a day I guess um okay let me So how do I share share

screen um yeah so you could see this all right couple clicks yeah give me a couple clicks couple clicks oh yeah yeah couple clicks to zoom oh now it's like looking at my mom's iPhone okay that's big um yeah so I've been working on uh MCP support uh native in Marra and yesterday y'all tried it out um but at the time we only supported uh HTTP streaming which

is the new protocol the deprecated one for uh remote MCP servers was like uh SSE and HTTP um but because it was wasn't deprecated that long ago most agents like uh you tried like claude and and cursor and maybe Windsurf uh it didn't support it uh they just don't have uh support for uh stream HTTP yet and so we

decided to uh add support for SSE so that's kind of what I want to show um do you want me to just show MCP in general i can't I honestly don't remember even what I showed yesterday show your MCP server that you made how you and you know all that yeah so let's Damn it swear I was doing pretty good with the conversation with David they didn't cuss that much yeah I think I think it's only $3 in a splur jar so far today so not

enough to buy me a drink later but we're getting there there's still time yeah maybe maybe like a PBR that's it um not in San Francisco uh so I've got a a couple MCP servers here uh they're not really doing anything anything that interesting uh I guess what the interesting part is uh you still write it with the same uh MCP server class that we had before but the difference now is that you can pass it

into your master instance uh so automatically gets uh telemetry added to it logging added to it uh it's kind of in your whole uh master ecosystem so wherever you deploy it that's deployed there as well um we also added some some routes to it uh to expose what I was saying before streamable HTTP and SSE as

well and so with that all I need to do I have my my MRA server running here and then inside MCP.json oh I didn't put it there I guess let's see yeah so I have my my server over here uh just demo MCP and I'm just pointing it at the the uh SSC endpoint that Monster gives me and so I could always check out the uh I'll pull up the Swagger UI as well so these are the the different

routes that it gives you so uh the old spec called for both of these routes but you only really have to be concerned about this one uh the new spec it just uses a single endpoint for uh all all the requests and so yeah I have it hooked up here uh let me ask what is the weather in San Francisco calls my tool runs a tool and bam 20 and sunny daniel's a cursor too yeah Daniel

you're a cursor cursor guy which is funny though Daniel because when I saw this demo earlier it was 20 and sunny where you're at right in Waterlue yeah I think Oh yeah let's test that again is it still 20 and sunny in Waterl uh what is the weather in 20 yeah what is 20 can you explain 20 room temperature yeah can you explain 20 for

the you know the Americans that do not know Celsius yeah dude most of the world understands what 20 means i'm not from Canadia yeah you got to figure it out it's not on me that's true okay we got we got to do this test you John where are you at it can't be sunny where you're at there too let's go with

Seattle seattle all right what there's It might be 20 degrees but it's not sunny it's also possible that this tool just returns this for every single thing it's not actually calling an LLM 20° 20° this is This is Wait wait daniel you you just did What is this actually doing demo peak fake demo right here oh my god so that it's not even calling

anything it's just defaulting i did yeah so it's So if it were calling out to an API a weather API don't we have an example that if you just create master it already has that baked in yeah yeah wanted to add some flavor here honestly I think I I think this I just told cursor to create a couple servers for me that do something silly there there you go 68 degrees is 20 but clearly we've

learned Daniel is cheating on his demos so you know he he he's vibe demoing as we're vibe coding here but we get the idea though it did call the function right it returned whatever you told it to return we could easily have that call out to an API or or do whatever there's this comedian I like who had this joke about

like Fahrenheit which is like makes more sense than Celsius because 68 Fahrenheit is 68% hot and 100 is 100% hot you know and 20 is like cold you know but I don't like you know that's that's makes more sense to me but in Celsius zero is freezing and 100 is boiling where where I'm from zero is way colder than freezing

i don't know i never seen snow fall so I don't know you've never seen not fall from the sky we got someone from Turkey Arie thanks for joining us i really like this feature where you can just like show stuff yeah we're just learning new stuff every day around how this works uh for all those of you that

are still joining us or that are showed up late we are Shane and Obby Daniel and Eugan from Mastra we've been told these are expensive meetings but we do them anyway and we had David from Centry on earlier showing us the Century MCP server we've been talking or MCP we've been talking A to uh we just saw a demo that was mostly real partially fake around uh how you

can build MCP servers with MRA yeah and yeah what do we got next we got weather was fake everything else was real yeah just the tool call everything else was was right it was pretty close Daniel thanks for coming thanks for joining thank you yeah have a good weekend yeah now we have you John to to to work with

and then there was just you John how's it going YJ pretty good um you know who you are yeah yeah yeah give a little insight into what who is uh Eug John yeah so um let's see uh well my name is E John i'm uh currently located in here in Seattle uh I grew up in Siri BC Canada and um let's see i really like coffee uh enough to pour all my money into that if I'm

not pouring my money into coffee I'm pouring it into my kid who's uh who's going to turn 10 months soon so life is busy where'd you work before in MRA i worked at Netlifi before here uh that's where I met uh yeah Shane and Obby and some other folks here at MRAA uh been working there for about two two almost three years actually uh like two years and 11

months nice you've been working a lot on our tracing and observability infrastructure i was hoping you could teach us about Clickhouse because we don't know what the that is oh yeah well I'm I'm still a beginner myself uh I can sort of explain why we were what was going on there um initially when we were

deploying uh Masha projects to cloud all the tracing was going to our livql uh database and periodically you know if there's enough traces being shoved into uh our livql database at once uh the database kind of froze that's no good um so we moved uh our tracing logic to go to click house uh which is a column

oriented database so we have a single table um where we store uh right now we're just storing yeah all our traces for the agents and workflows in one table and we sort of uh you know have the ability to filter them off by your agent name or your workflow name and we've seen some pretty good performance now and we haven't ran into those issues fingers crossed but it's looking good

finger good here nice dude what's your favorite part about working at at Mustra the favorite part there's lots but um I'll say I'd say that the work I'm doing is really interesting um I'm I've always been interested in kind of cloud stuff like we were I was doing back at Netifi but this has another flavor to it an AI you know an AI flavor uh before before

joining Monster I wasn't really into AI but now you know I'm ho I hopped on the bag bandwagon and know everything's AI now like like all of us we we are buying into the hype right right um I'm team Win Surf by the way uh that's another fact about me right all right so yeah so we're like 50 We're like 5050 represent yeah i just don't like dragging my you

know files every time i like I just want Windsurf to just you know take it away you like to go surfing nothing wrong with that surfing is where it's at um cool dude well thanks for joining us for Friendsgiving right here have a great weekend dude yeah we'll see you later see you John all right everybody thanks for everyone

who joined we just hit 200 took took like 50 minutes but you know we got 200 people watching seven minutes to go so to kind of do a a quick recap of what we've covered so far we talked a little bit about MCP we had a guest David from Sentry come on show us the Century MCP server that he's been working on he also has a whole bunch of hot takes on MCP on his blog so we should find that do you

have that link maybe we can pull that up um we'll uh we'll find if you're looking for some I would call it not light reading because they are pretty long but they are very detailed so I would actually you know if you're interested on some hot takes on AI or MCP specifically check out some of the things he has been uh he has been posting there and give him a follow of

course right there easy enough uh to remember there and we also then talked a little bit about ADA we had some people from Mastra join show how you can actually deploy your own MCP servers with MRA which is cool and yeah learned a little bit about uh Eugen and some clickout stuff anything else you want to do before we

wrap up today dude i don't know i think what are we doing tonight oh we have an event tonight right yeah yeah let's talk about it yeah yeah so we're going so there's an event at the YC office tonight with a whole bunch of you know YC companies YC alums invite only sorry yeah sorry sorry you know but we'll be

doing a de a demo there so that'll be cool whole bunch of dev tools companies superbase yeah we're co-hosting with Superbase starling Chon yeah a whole bunch of demos actually you know what i'm gonna shout them all out because that's what you do on live streams right yeah let's Let's do that these are all the homies for tonight can you show your

screen let's just Oh should I share my screen no just shout just shout them out yeah we got Super Base Mra Starling Chunky Clavis AI Vibe spelled with a Y v Y B vibe vibe delt ai better off that seems needed better dude better off i think better off was at the vibe uh hackathon that James did but anyway um run RL Prism AI Morphic guess what food

we're eating tonight guess just guess i'm going to guess if you're in San Francisco and you're at a meetup or an event 95% of the time you are eating pizza pizza dude it's always pizza it's always pizza you this is a hack for for any of you if you want free pizza just go to Luma search AI find an AI meetup there's an AI meetup every night of the week and you can eat free pizza that is just you know guaranteed

guaranteed i went to a Cloudflare meetup once and I saw a girl who will not be named because I don't know who she is she went there she had five slices of pizza and then dipped and I was like dang that's smart so we I mean we hosted an event where someone not in tech came talked to some people grabbed some pizza

invited her boyfriend and then left as well so it's it's a thing i mean I don't think I don't think the companies or VCs care or they are not enough people have caught on yet where it's a problem but if you want free pizza go to San Francisco all you can eat yeah Luma is your pizza gateway but the pizza will be

trash though most most likely yeah not always you You can find some good pizza we had some good pizza earlier this week on Monday if you come to our events the pizza will be trash i'll tell you that right now so don't expect too much okay long bridge shout out Longbridge probably wait do we have to order that still yeah we have to order oh we didn't even order okay well we'll do

that we'll we we'll try to get the some good pizza uh we're going to do this quite a bit uh try to live stream pretty frequently so follow us on here should tell them tell them about Yeah uh yeah so many of many of you all are from Japan um and um are developing with MSRA we have a event in Japan on May 19th uh it's at the all Allstar SAS um it's in

Ginsza Tokyo i've never been so I'll be there tony will be there if you are Japanese in in Japan or you live you don't have to be Japanese if you live in Japan uh if you want to meet us there's going to be some events around there um and then you know Dave we need to get you hey Dave do you want Do you want to come on right now dave one last phone a friend i mean we can go a little long

dave if you're if you're around and you just want to jump in get on dude get on we'll send it to you yeah Dave we'll we'll send it to you if you can make it we'll stick around for a few more minutes and we can talk about some of the stuff you've been building i know you've been playing around with MRA

um but yeah that I mean sure that seems like a two games on one day yeah it's it's a double Dave day but yeah make sure you follow MRA on Twitter mostra AI follow me on Twitter small he's down all right cool um he cool he's in he's in a Whimo all right oh that's perfect this is like mobile phone a friend yeah now

how do we get I sent it to you on Slack Dave if you're if you're listening yeah if you can get on Slack jump on in and we can keep chatting also you know I'm about to ask you for pizza right yeah did I think Dave is that Did you volunteer did you just volunteer i hope we're seeing you later too yeah i hope you hope you volunteered

there he is yes what's up what's up guys thanks for having me on yeah phone phone a friend literally phone a friend while you're drive while you're in a vehicle i'm I'm on the way to YSP right now i'll be at the event tonight oh nice yeah so was that was that you telling us that your investor you know investor should order the pizza

you're going to help us or what yeah yeah should I write down the order for you guys right now and I'll call in no no no no we can take care of it but um so you've been playing around with MRA a little bit right i have yeah most it's it's been a game changer um I mean it's it's everything I've wanted for like so so we build a lot of our internal software and our tech stack is is Nex.js

so obviously TypeScript is huge for us um and it was really awesome because I was able just to turn a bunch of server actions that I already had built out in my application just into tools that the master agent could use um and I I was on like an older version of Master so I need to get on uh V- Next and and

totally refactor my code now sorry about that but um but no I I actually have a a pretty large file that is basically just extracting all of the companies from YC's website for the current batch um and we use it so that we can track who's in the batch which companies are launching who's working on what um and just some background Orange Collective

we're an investor in Maestra we invest in a ton of YC companies and so for us that's pretty critical um and right now I just have like a basic we we use Firecrawl on the back end to help with the the extraction um and it's just this long really kind of hairy file that to be honest I just viodated so I need to refactor this into like a multi-step workflow and I've

actually been going through the file to break out step by step all the different um I guess all the different steps in that gigantic file that I can then turn into uh a mantra workflow with with V next and so sign up for the hackathon next week because because I get I gave myself a little assignment to to turn it into a a master agent next week you should come

on the show and we can build it together let's do it be awesome yeah yeah we we can help you build uh you also mentioned Firecrawl we know some people over there but you know chat help us out let's get someone from Firecrawl on here we should bring have them show off some of their stuff that they they're working on some

really cool stuff over there too firecrawl is amazing um you know you can define your schema very similar to um you know in in Maestro you can als you guys also use Zod right for um I I guess defining certain parts of the tool chain but um with with fire crawl you just define the schema that you want to extract out of a website um and you give

it natural language prompts and it uses a combination of you know LLMs to extract uh uh plain text from the side but also parse that markdown file or HTML file it gets rid of all the nasty headers so it doesn't totally um kill you on token consumption um and so they do a bunch of optimizations under the hood that quite frankly I don't even know exactly what's

going on but um it's great that it's all abstracted for me and it's kind of a great fit with with Mustra awesome dude thanks so much for the impromptu join we'll see you later today yeah we'll see you later dude i like doing this live stream stuff if like the the homies can just jump in from like car everything yeah phone a

friend is real phone a friend is real uh thanks everyone for joining you know just to recap you know talked about MCP today all this stuff's recorded too so if you ever go to our YouTube it's on the MRA Twitter as well so if you didn't catch it if you came in late if you want to see you know specifically some of the stuff at the beginning was really interesting watching you know David from

Sentry talk through how they're thinking about MCP over at Sentry how the MCP server that they built seeing the demo seeing how a lot of this stuff still like it works but it doesn't always work right we're struggling with it just like you all are uh but yeah appreciate you all being here yeah thank you so much any parting words