my name is David Andre and here's everything you need to know about the A2A protocol So what even is A2A This stands for agentto agent protocol a new standard released by Google The big issue with AI agents is that they are all over the place You know from different frameworks companies APIs tools they're just everywhere A2A fixes this This is the future of AI agents So in this video I'll show you what A2A is and how to use it yourself Real quick I'm hiring a back-end developer to help me build Vectal So if you have five plus years of experience make sure to apply the link is below the video Okay So here is a nice visual way to understand the A2A protocol So this is you or me the user and this is our main AI agent right The one you built for your app for example Now with A2A this client agent can easily communicate with any remote agent So from other companies other tools from other frameworks it doesn't matter If the agents are compatible with the A2A protocol they can all communicate with each other easily and quickly Now here's why I believe that this is the skill of the future Everyone always asks me "David what's the next big thing?" Well A2A is that thing Learning how this protocol works will give you such a huge advantage over everybody else Think about this Most people still don't even know what MCPS are or what they do or how to use them That's why learning how A2A works will easily put you in the top 1% of AI So watch until the end You can think of A2A as a common language for AI agents The same way English allows us humans from all corners of the world to communicate easily the same way A2A allows AI agents from all different codes frameworks companies to communicate with each other And the setup is super easy Each agent just exposes one HTTP endpoint and a small JSON card So basically A2A makes any AI agent you build 100 times more scalable and future proof than before Now just to give you a taste of the future here are all the companies contributing to the agent to agent protocol And again this protocol was developed by Google So let's say you build a new AI agent right and you make it compatible with the A2A protocol Your agent can call Jet Brains or Coher or Deoid and use their own AI agents and leverage whatever AI agents they have built So even though we are in the very early days of this protocol it doesn't take a genius to imagine just how powerful this might be very very soon Now here is how A2A differs from MCP MCP makes connecting tools and data to your AI agents so much easier right So here's an example of MCP You have a source agent then you have the MCP server and then you have all these different tools Could be a database could be all kinds of APIs the web anything So MCP makes connecting tools and data easy While A2A on the other hand is about connecting other agents with your agents So again you have your own AI agent here and then in the middle is the A2A protocol which allows your agent to communicate with any other agent using the protocol So it's important that you understand that these two protocols A2A and MCP do not compete with each other Instead they complement each other very nicely making agents easier to build and distribute Here's another visual way to understand how MCP and A2A can work together But shout out to the people who made it because this is a great graphic So you have your agent right here Again your main AI agent using the A2A protocol it can contact other AI agents Right now in the middle you can have your MCP server which allows your AI agent through the MCP client to get all kinds of prompts resources tools anything else compatible with MCPS So looking at this animation you can understand how making your AI agent compatible with these protocols like MCP or A2A can make it a lot more powerful and useful By the way if you want me to make more videos on the A2A protocol then make sure to subscribe Currently only 21% of you watching are subscribed So please go below the video and click the subscribe button It helps out a lot Okay So if you want to really understand A2A there are four concepts you should wrap your head around And the first one is the agent card Think of this as the agent's business card but written in JSON which stands for JavaScript object notation The agent card announces who the agent is what it can do and where and how to talk to it So the agent card lets your app automatically discover the right agent for the job The second concept you need to understand is the A2A server This is the live bot running on your network ready to work It listens for incoming requests does the work and then sends back the results or asks follow-up questions if needed Think of the A2A server as the engine that's actually processing the commands in a consistent and predictable way The third key concept is the A2A client This could be any program or even another agent that wants something done The A2A client reads the agent card packages your requests into a task sends it off and collects the answer Basically the A2A client acts as the bridge between your users or your systems and the agents So you don't have to write custom code for each agent which is kind of the core idea behind having this uniform standard that everybody uses Now before we get to building there's one more concept you should understand and that's the A2A task But don't worry this one is the simplest A task is basically a single to-do that you hand off to your agent It carries your request through its entire life cycle Submitted in progress finished The A2A task provides a clean uniform way to track and manage every little job every single prompt that you give to your agent Now here's why I think we're at the right place at the right time If you can't already tell this is a major breakthrough in the field of AI agents And yet nobody is talking about it Why Because it's not flashy you know It's not vibe coding It's not building a crazy website in a matter of a single prone It's a bit more complex right But that's why getting into it will yield even more benefits because the competition is that much smaller To put this into perspective imagine reading about the TCP IP protocol aka the protocol that the entire internet runs upon the same week that it came out This is the week that A2A came out With that being said let's get to building Okay so here's the GitHub repo which of course I'll link below And what I'm going to try to build and show you in this video is an A2A B2B MCP AI agent Now if you think I'm just talking no this is an actual term and I'm going to break it down in this brilliant graphic that I drew myself By the way this is copyrighted work So if you steal this you're getting sued Anyways you can see the color coding right So A2A stands for agentto agent protocol And this is the way the different AI agents communicate with each other Now B2B stands for businessto business right This is just a product or a service that's being sold to other businesses So here you can see the squares are different companies different businesses MCP stands for the model context protocol This again allows AI agents to easily access other tools and data So in my graphic right here green is AI agents right So these circles are different agents within these different companies As you can see each agent can access different MCPS aka different tools Now thanks to the A2A protocol all of these businesses can communicate with each other with their AI agents So instead of using slow and outdated methods like email which people only respond to 9 to5 and then not on the weekend you can have your AI agents running 24/7 communicating with agents from other companies So if your company is doing a project with another company and both of you have AI agents on the A2A protocol you can have your work being done while you're on the toilet or while you're laying on the beach even if the other company's CEO is laying on the beach as well This is what the future will be like Most people will don't get it Most people ignore it and just you know watch Netflix But if you really think about this if you spend you know if you go on a walk later today and really try to think like what will the future of business and the future of AI agents look like it will be something like this So let's get started with this A2A repo which again this is an open-source repository from Google So it's just released and it's like over 12,000 stars already meaning it's becoming real popular real quick So I'm going to show you how you can use this repository to build your own AI agents using the A2A protocol Okay so let me switch to cursor You can use wind surf VS code doesn't matter Let's open a empty folder I have this empty folder named code And boom there it is Now the very first thing we need to do is we need to clone the repository So either press command J if you're on Mac or control J or click on terminal and new terminal Basically we need to open the terminal Then go back to the repository and we need to clone this repo So click on code clone So click on code and then copy this URL Switch back into cursor and type in get clone and paste it in Boom This should clone the entire repository into your local setup And on the right we can see we have the folder or actually you will have it on the left Open the A2A folder and in here we have everything from the repo So should have all of the same folders and all of the same files Beautiful Okay let me type in clear to clear the terminal The next step we need to CD into the A2A folder So CD A2A Enter Real quick if you want to be at the cutting edge of AI then make sure to try Veectal Veal is an all-in-one productivity app that actually saves you time For example a new feature that we added is the background agent So let's say you want to get started on this task where you can simply click on activate vectoral agent and just like that the AI agents will start reasoning browsing the web thinking and just working on your task autonomously in the background without you having to lift a finger And if you have vectal pro you can activate this agent for all of your tasks on the list no matter how many tasks you have Plus Veal Pro users can access all of the cuttingedge AI models available today So while most people still use outdated productivity apps like ClickUp or To-Doist those who see the future and those who like to be hyperproductive are quickly switching to Vectal So go to vectal.ai and give it a try I promise you won't look back Next up we should activate a cond environment So I'm going to list out all my environments Uh I already have one for A2A but if you don't you can simply ask how do I create a new cond Just answer So I'm going to do cond activate activate A2A Boom Clear So yeah this is how you create a cond environment Super easy If you don't know how to install cond This is the applicable for any point in this video If you get stuck at any point if something isn't clear take a screenshot and ask CH GBT Actually I'm going to show you how I work with the new 03 model because the new 03 is really OP So I'm going to show you how I use it to work because um yeah having access to O3 is really illegal I mean it feels illegal It's absolutely illegal but yeah for $20 a month obviously I'm paying 200 for the pro plan but even on the plus plan using 03 with all the tools it has man amazing Anyways let's focus up on A2A Let's open the terminal again So we are in the cond environment That's good So actually the way to check that you can just open up any you know python file and then in the bottom right make sure you activate the cond environment you selected so that then any terminal you create will be in this environment right a2a that's important okay let's see the a2a again and next up we have to run this command cond install- c cond forge uv uv is a python package manager that's way faster than pip so we're going to be utilizing that in this video now in the meantime time while this is installing let's go to the right and let's click on samples Python and you can see the sample AI agents made by Google you know the ones that they offered in this repository that already are compatible with A2A so we can click on agents and we have CI which obviously is one of the most famous frameworks we have langraph we have Google ADK which if you want me to make a video on Google ADK comment below we have some other ones but uh for this video I'm going to showcase how these two can communicate with each other so this is the agent for QI And this is the agent for ADK So I'm going to adjust both of these Actually I'm going to use cursor for that So let me expand it I'm gonna say I'm going to add files to context Agent create ADK Actually I'm going to update these files by adding a comment Google ADK agent Okay So now when I open a new chat in cursor I can just type slash add and this will add all open files to context This is very important super OP shortcut If you're not using it you're missing out What I'm going to say is update the prompts of both of these agent files so that the Crew AI agent gives concise and clear responses but the Google ADK agent always gives thorough and detailed reasoning for any prompt we give it So again just to reiterate agent to agent protocol is building for the future You're basically betting that AI agents will be big which I guess all of us already believe that AI agents are the future But the main thing that's different here is that instead of choosing the right framework which again that's another common question I get Which framework should I use to build AI agents David it doesn't matter now It literally doesn't matter As long as you make your AI agents compatible with A2A you can have your agents interact with other agents from any other framework So this is what I'm going to be setting up here in this video So pay attention Okay So Gemini did not update the prompts That's crazy I'm going say you did not do anything Update the prompts Do not change anything else Sometimes this happens where Gemini 2.5 Pro is lazy and just doesn't do the actions When that happens let's just switch to CL 3.7 Max which doesn't have this problem I don't know why Okay so there it is It's updating the first file Okay so yeah let's accept And now the Crea agent Okay So yeah it just reroll the prompt Looks good Accept All right So the next step is to run the samples Let's open our terminal again Uh this has finished running All requested packages already installed Amazing So let's clear it And then we need to cd into this folder right So samples Python agent CI Boom CD samples/python agent CI So you need to be in this folder in the CI folder Okay So next up we need to get the Google API key So go to your browser type in Google AI studio log in with your main Google account and then click on get API key Now again treat API keys as passwords Do not share them with anybody So click on create API key right here Now here you need to select the Google cloud project If you don't have one super easy Takes like 30 seconds to create one And then simply click on create API key And takes two seconds Boom Copy it Again I I will delete my key before uploading this video But feel free to try it if you have extra time Okay So once you copy this API key go back into cursor and we need to export it into environment variable So pay attention Type in echo quotation marks Google API key equals and then paste in your key and quotation marks greater than Envoom hit this command What this should do is create a new environment file in this create folder with the API key already attached Now if you're scared of terminal you can literally just do that Click on the folder new file.env envi key equals and then paste in your API key Then we can clear the terminal again and type in uv run dot enter So this is starting the create agent Now while this is starting we can create a new terminal and we're going to start our Google ad agent which is this one And actually here I'm going to show you the other method So just click on the folder click on new file envi equals and paste in the API key So inside of this new terminal we need to CD into the A2A folder again Again which is following our structure on the right Code is my folder that I created A2A is the main folder from the GitHub repo that we cloned right A2A So now we need to get into this Google ADK folder which we will achieve by running this command CD samples Python agents Google ADK Boom And you should see the folder name right here Let's clear the terminal And since we already added the environment file manually we can just do UV run dot Boom And this is starting the Google ADK agent So now we have Ca agent running and we have the Google ADK agent running Now to test that you can open the browser open a new tab and paste in the URL of the local host server that the agents are running on So CI is um 10,0001 and agent SDK is 10,0002 Now you have to do wellknown/agent.json JSON to see the JSON of this agent So this is the image generator agent description generates stunning high quality images on demand blah blah blah And if we do the other one this is the reimbursement agent This agent handles the reimbursement process on the employees Okay so as you can see this is just a demo example from Google But if you have the slightest bit of imagination you can see that this could be any AI agents from any frameworks any companies communicating with each other using this protocol Okay Okay So then I'm going to open a third terminal and I'm going to CD into A2A again And then I'm going to CD into this demo folder And actually we need to CD into demo NUI Demo/ UI Okay This will allow us to use a nice user interface that Google has made And on the UI folder right click and type inv.local And in here we need to assign next public agent URL And then we're going to do the first one which is on localhost 10,0001 in the terminal then do npm install Now actually I'm going to add a second API key here the Google API key and I'm going to attach the same value we already created We don't have to create a new one And then in the terminal type in uv run main py This should run the Python UI Okay Installed 98 packages in 362 millconds That's the power of UV Anyways it starts a server on this URL So let's click that open And this is what it looks like This is the UI Google has made Super simple It asks for the API key again So I'm going to provide it again Save Let me zoom in a bit Let me open the chat I mean the sidebar So we need to add our agents If you click on agents they're not here So we need to click is agent address And these are the two addresses we have right here So literally copy this agent address in here Read Actually no Uh all we need to do is the local host We don't need the full thing Just the local host Just the address Read Let's check if our agents are still running They are I don't know why this is getting rejected Okay Let me just do local host without the HTTP read Oh there it is Okay So you have to do local host colon 10,01 So this is the image generator one This is running on CI Okay The UI is kind of sketchy Google I think someone at Google needs to vibe code a better UI Um yeah they didn't do a good job but again it just released All of this will get much better The main thing you have to understand is the implications of this standard like something like this needed to exist because there was so much chaos in the agent space So that's the main thing All the details will get polished out I mean you know the first version of CHBT was pretty shitty compared to today So think about this The first version of A2A Anyways I'm going to save this agent And now we have the first agent in here We need to add the second one So let's do that Boom Let's click this upload button again Add the second one but without all this stuff Just local host 10,0002 Read There he is Reimbursement agent Save So we have these two agents added So if we go home we can click on new conversation and we can say "Hi what AI agents are available to you?" Enter Okay Okay I have access to the following AI agents There it is Image generator and then reimbursement So I'm going to say generate an image of a green flying cat Now obviously these AI agents could be much better and this is just a demo Do not be fooled by this like "Oh my god David why haven't you built a AI agent that automates your entire business?" Obviously ideally we would build a super useful team of agents But the main point of this video is to understand the A2A protocol Every single video on this so far I've seen completely avoided any building They didn't show how to get it up and running Didn't show how to clone the repo how to set it up They just like read the basics and then avoided all the difficult part That's the difference I wanted to show you guys the difficult stuff right We ran into a bunch of errors but we solved all of them and now we have this UI running with these two agents active Um I don't know why this is stuck Maybe it's generating the image Not quite sure By the way you can change the frequency of the HTTP request because 1 second is kind of brutal If you look into the terminal it's pulling it every single second So this is a lot of requests Um I'm not sure if you want that If not just change it to 5 seconds That's a good balance I think Anyways this is the A2A protocol Obviously the UI will improve All of that will improve The AI agents will get more useful But the main point is that no matter what framework you use that's the main thing you have to take away from this We have one AI agent which is um right here running on CI second one running on Google ADK and they can communicate with each other through A2A protocol That's the main thing that's the breakthrough that Google has managed to pull off with this protocol And again you can also use langraph lama index marvin semantic kernel any other you know openai um swarm or openi agents SDK anything that you can connect to the A2A protocol which again if you want to learn more about this visit the official repo I'm going to link it below the video of course from Google definitely read through this it's you know honestly I could make multiple videos on this and if you want me to make more videos on the A2A protocol then comment below because again there's a lot to We haven't even gone into the more advanced concepts We just went through agent card ATA server ATA client and task But yeah there's a lot more to this and again this is the future of AI agent So if you're able to just wrap your head around this and understand that this is the starting block This is day zero This is month zero Okay month zero of the A2A protocol It's only going to get better only going to get easier to use The UI is going to be nicer All of that stuff Don't worry about the details Don't worry about the errors Look at the big picture Look three months six months 12 months into future and imagine okay the pace the just the last three months the pace of advancement has been crazy right Even three months ago nobody was really using MCP I mean my first video on MCP was four months ago but that's because I was early right So those of you guys who watch this channel are also going to be early to A2A and other future changing technologies that are to come So again if you want me to make a video on A2A another video on A2A perhaps going more in depth or building a more advanced team of agents or some other idea just comment below And also if you are using any task manager or productivity app just switch to vectal okay You're going to save yourself so many hours going forward And if you aren't using any task management tool you probably should You should have a to-do list somewhere And if you're going to put it anywhere Vector is the perfect place because we have advanced AI agents running on some of the best AI models in the world inside of Veectal So again go to vector.ai sign up It's completely free But if you're a serious user upgrade to Vector Pro get access to all the best models get access to the background region to infinite thinking to Ultra Search and everything else we have to offer So again go to Vectorai give it a shot With that being said thank you guys for watching Hopefully you learned something new and I see you next time See you