results/transcript_analysis_1745932831.md

Transcript Analysis

Summary

• The video introduces Google's Agent2Agent (A2A) protocol, designed for AI agents to communicate and collaborate.
• A2A enables secure information exchange and coordinated actions between agents across different platforms.
• The video contrasts A2A with the Model Context Protocol (MCP), explaining that MCP allows an agent to access external services, while A2A facilitates agent-to-agent communication.
• Agent discovery, facilitated by configuration files (agent.json), allows agents to identify and interact with other available agents.
• The presenter provides a coding demo showcasing how agents can interact using A2A, including an example of currency conversion.
• Client agents initiate communication and manage tasks, while remote agents provide specific functionalities or access to data.
• Both A2A and MCP are evolving protocols crucial for the development of agentic AI applications, requiring adherence from service providers and collaborators.

Topics

• Agent2Agent (A2A) Protocol
• Model Context Protocol (MCP)
• Agent Communication and Collaboration
• Agent Discovery
• Client and Remote Agents

Notable Quotes

Obviously you can understand what A2A protocol basically does. It it actually helps you to perform some kind of communication between agents to agents to perform some kind of task.

A2A is an open protocol that complements anthropics model context protocol.

MCP is nothing but model context protocol. It is a protocol which is specifically used by an agent or an AI assistant to probably get some kind of context from this external tool providers or service providers

MCP will be available within a specific AI assistant... But agents altogether can be considered combining this MCP with the AI assistant and this will basically communicate with the other agents itself.

Now see guys, don't worry about so many developments happening in this specific field. These are still evolving. MCP is still evolving. A2A is evolving.

Formatted Transcript

Hello guys. So in this video we are going to discuss about this new protocol which is called as agentto aagent protocol which has been launched by Google. We also say it as A2A protocol and uh if you remember like uh recently uh anthropic has also come up with this model context protocol itself. Uh so obviously Google will not be behind you know they'll also come up with something. So that is what they have actually come up with agent to agent protocol. And uh in this specific video we are going to talk about this. We're going to understand how it specifically works. And I'll also be talking about the agenda of this particular video. Right over here you'll be able to see this particular diagram. Uh here there are agents multiple agents and all the agents are basically connected. So obviously you can understand what A2A protocol basically does. It it actually helps you to perform some kind of communication between agents to agents to perform some kind of task. Okay. So going to the agenda what all things we are basically going to discuss in this specific video. So first of all we'll start what is A2A protocol agent to agent protocol how does A2A protocol work difference between A2A versus MCP model context protocol because we also need to understand this you know uh like how this two specific A2A protocol and MCP protocol differ and then we will also be seeing some kind of demo with coding where we'll run this kind of we'll see how this A2A agent communication will basically happen and then I'll also talk about the future thoughts uh with respect to A2A protocol call and MCP protocol. So step by step we will be seeing all the specific points. It is important for you all to understand because this protocol will be really necessary as we go ahead and since this agentic AI application completely is evolving this will be very much important both MCP and A2A. Okay. So it is necessary that you please watch this video till the end. Okay. So first of all uh let's go back to the docu documentation here. You can I'll just read some of the lines and then with basic examples I'll try to showcase each and everything. Okay. So AI agents offers a unique opportunity to help people to be more productive by autonomously handling many daily recurring or complex tasks. This we already know right? If you're building some agentic AI applications there you'll be able to see that AI agents will be able to perform many many tasks. Even AI agents will also be able to communicate with the other AI agents to solve some complex task itself. Okay. Now they are launching a open protocol which is called as agent to agent with support and contribution from more than 50 technological partners. So Google when they came up with this thing um they have also a lot of technological partners that is good you know like bigger companies like Atlaxian, Box, Coair, Intude, Langchain, MongoDB, PayPal, Salesforce, SAP, Service Now, UK uh UKG and Workday right and then lot of other companies who and this A2A protocol will allow AI agents to communicate with each other securely exchange information coordinate actions on top of various enterprise platform or applications. So what this is basically doing is that this protocol will actually help one AI agent to communicate with the other AI agents. Now just imagine if some of the functionalities from this particular company is implemented in a form of agents and that is exposed to the other agents then the other agent will be able to collaborate and perform various task itself. Again this is just a thought process but I will show you multiple examples. I will probably draw each and everything for this. Okay. Here also it is written A2A is an open protocol that complements anthropics model context protocol. As I said that it is just like complementing on top of MCP. Okay. So we'll also understand the difference. Now let me go back to my uh coding screen and over here you'll be able to see writing screen. Sorry here I have probably given the definition. It is an open protocol that complements anthropic model context protocol which provides helpful tools and context to agents. Drawing on Google internal expertise in scaling agentic system. We designed the A2A protocol to address the challenge we identified in deploying last scale multi-ere system for our uh customers. Now to understand what is A2A protocol. Okay. So let's say that I have this as one of my agent and this is my another agent. Okay. Let's say right now u if this is my agent A. Okay. So I will just go ahead and write this is my agent A. Okay. So let's say this is my agent A and this is my another agent that is B. Okay. Let's say this specific company you know is just like a booking website. Okay. Booking website. One of the booking website that we commonly use in India. Let's say as an example it is make my trip. Okay, it can be make my trip. It can be a it can be various other sites you know it is up to us like whatever site you specifically refer. Now let's say that the input given by the user is that hey please travel my plan or plan my travel. Sorry I'm not saying travel my plan my travel. Okay. So, plan my travel. Okay. To some countries. Let's say I want to probably go to Europe. Okay. For 7 days. And here we will say that one of the important goal is with minimum cost. Okay. With minimum cost. So I tell this specific agent, this agent is of this particular booking website and I'm telling them to perform this specific task. Okay. Now what agent A will basically do, it will go ahead and communicate, right? It will probably go ahead and communicate with different types of agents. This is my agent A B. There may be other agents over here. There may be one more agent somewhere here. Right? and this agents just imagine I'll go ahead and write this is my agent B then let's say this is this is my agent other agent over here okay please try to understand this okay so this is my agent uh C this is my agent D this is my agent E okay okay now with respect to this particular plan travel first of all we need to see that what are the intermediate steps okay first of all we think that we need to book flights right so obviously I need to probably go ahead and communicate with an agent which is responsible in booking flight right so here I will be having this specific agent and this agent will be developed by a flight booking company let's say that this is nothing but Emirates okay let's say this is Emirates Emirates okay Emirates is one flight booking industry right so this agent will first of all try to find out that hey whom do I need to probably communicate with for booking flights? Okay. And let's say this agent A will be able to identify there are two different agents C and E which will be which are nothing but these are agents of some other companies. Okay. Some other booking flight companies. So here is Emirates. Let's say this is another flight which is like Air India. Okay. Air India. So Emirates and Air India will be able to probably tell give some kind of information with respect to the price. Okay, with respect to the price. So let's say that over here I am getting some kind of information from Emirates. Hey, this is what is the price that I will be getting since uh these are the agents that is probably communicating over here. And remember this Emirates will have internal data information, right? And this Emirates, this agent that is created, it will be specifically for this Emirates. This agent E will be specifically an agent for Air India, right? And this is how this agent is able to identify it is able to take the information from both of these particular agents and then it'll get a kind of response over here and this response will be having whichever will be the minimal cost it will take that specific options and obviously from Emirates and Air India I feel Air India will give us a affordable cost with respect to the flight travels. Okay. Now the second thing is that we need to book we need to book hotels. Okay, we need to book hotels. Now for this hotels let's say that I have B and D agents available right and how it'll identify which all agents are available that we specifically called as agent discovery that also I will talk about it. Okay. So through agent discovery how this agent discovery is done I will also talk about it. But let's say that from this agent discovery. So here we basically go ahead and discover agents and we will get to know that how many different types of agents are there. Now let's say based on this agent discovery we know that for booking FL hotels there are two different kind of agents available. Okay let's say this is one this is the other one. Okay let's say this agent is specifically belonging to Taj hotels. Okay I'm just taking some example and this agent is basically belonging to let's say Marriott. Okay, variate. Now if these two agents are available, so what this agent will do is that it'll try to communicate with them and based on this this will try to get some kind of information and whichever is the affordable one, it will select those. Right? So this is how agentto agent communication will specifically happen. So now similarly you can probably consider any kind of different kind of use case uh considering a aggregator platform you know. So let's say make my trip is one kind of aggregator platform which will actually help you to connect to flights, hotels, um you know even probably plan the entire trip in short right similarly you can take another example like let's say for jobs you know for jobs in India we definitely have lot of job portals right so there are job portals now this job portals can basically create an agent and connect to various companies agent which are responsible in hiring things you know hiring hiring candidates candidates right and based on that right let's say if Dell if uh there's a company like IBM there's a company like uh HCL there's a company like Sapient if they want a specific requirement what they can do is that that specific agent can probably communicate with this job portal agent and probably provide uh candidates with respect to that right uh similarly let's say that over here I have this particular job portal again the communication will happen in that same way right agentto agent communication that what exactly A2A basically means. Okay. Now if I go further right how A2A works I've already spoken about it you know by drawing all the specific diagrams but if I talk about architecture point of view there are two types of agent one is the remote agent and once one is a client agent. So this agent what we are speaking speaking about is it is basically called as client agent. Okay. So this is nothing but it is it is your client agent. Client agent. Okay. And these all agents that you'll be seeing it is basically nothing but it is a remote agent. Okay. This is nothing but remote agent. Right. How client agent is able to see so many agents? It is from this agent discovery mode. Okay. Through the agent discovery, this client will be able to understand how many different agents are available. Right? I will talk more about this agent discovery. how this agent discovery is you know there is some configuration files we need to probably set it up for this okay every agent needs to probably set a configuration file for this and we'll talk about that particular file and this file will be exposed to all the other agents now this agent is communicating agents now this agent is communicating to this right this agent can also communicate to this agent this agent can also communicate to this agent this agent can also communicate to this agent right this agent can also communicate to this agent right this agent C to D also can communicate C to E also can communicate so how every agent will know that there is some other agent because of this configuration file uh settings you know and we'll talk about that particular configurations file once we probably do the demo in in respect to coding right now let's understand this so there is a client agent there is a remote agent uh these all are specifically remote agent the client agent has some more functionalities you know all the communication that basically happens between the client agent and remote agent will have the secure collaboration it'll make sure to manage the task and state management user uh experience negotiation And even capability discovery as I said capability discovery is like the client agent will know that how many remote agents are available and this is through some kind of configuration configuration. Okay. And uh this configuration uh will be nothing but it is something like it will be exposing the entire agent card. Okay. We basically say it as agent card. I will talk about it as we go ahead. Okay. So this gives a clear idea about how A2A specifically works. But now one question that arises what is the difference between M2 MCP versus A2A right? Now let's say that I have this agents right so I have this agent one I have this agent two right I have this agent three right let's say these all are my agents now within this particular agent let's say uh let's say this is my agent A B C right and with respect to MCP also let's say I will go ahead and select a circle now agent to agent communication is definitely happening using a A2A protocol but when we talk about MC MCP MCP basically means what? Let's say this specific agent requires some of the services that needs to be required from a service provider. Let's say this is my service A, service B, this is service C, right? And let's say this agent wants to specifically use this service. Okay. Oops. Just a second. Okay. So this is my agent A and let's say this agent A wants to access uh is connected to various services and uh based on the services it requires some kind of information let's say this is internet search let's say this particular service is about it's uh DB connection local DB connection it can be third party APIs anything as such right so this agent A is basically communicating with the specific services with the help of MCP P the model context protocol and obviously if you remember in my previous video I've also discussed about MCP right and similarly this is my agent B okay agent B now inside this agent B let's say that I have different different different uh uh a uh services that I really want to use over here also so let's say this is my service A service B and this is my service C okay so different different agents uh will basically be connected over here again with the help of MCP and this will basically be doing this kind of communication and you know that what is MCP? MCP is nothing but model context protocol. It is a protocol which is specifically used by an agent or an AI assistant. I can also say AI assistant to probably get some kind of context from this external tool providers or service providers, right? So if I probably consider this this can entirely become okay one agent. So let's go ahead and draw this diagram also so that you'll be able to understand it very much clearly. So this can be my entire one kind of agent and internally an agent can have many MCP connections with respect to different MCP servers. Right? So this internally agent can have different different connections to the service provider right and this will happen internally to this particular agent all the specific communication will happen when we are talking about agentto agent communication that basically means this agent is communicating to the other agent right but it will not have any idea like what all MCP servers this will basically be connected to we are just connecting this agent to agent communication to get the information or work done by this specific specific agent right similarly for this specific agent also we are doing in that similar way right but internally this agents man can have any number of MCP connections okay so just to show you one very good diagram so let's say that this is one of my agent let's say this is one of my agent let's say this is one of my agent I can I can probably change this specific diagram to something like this let's say here I have one I have one MC uh one MCP server two MCP server and this is my AI assistant or agent itself Okay, this is my AI assistant or agent that is communicating to this with the help of MCP. Okay. And this is entirely my agent. One agent. Let's say this is agent A. This agent A is communicating to agent B to get the information. But internally this may be connected to some kind of MCPS also. Right. So MCP and A2A. I hope you have understood the basic difference. Okay. Now the question rises Kish can you show some example? And we definitely need to see some kind of demo with coding. Right. So for this what I've done is that I've directly taken the entire repository that has been given by Google. Okay. The first thing is that we'll try to understand how this discovery will basically happen. How this agent discovery will basically happen. I told you that we are going to set up some configuration file and even what Google has basically done is that whenever you create any kind of uh agents there will be one JSON file on this name wellknown agent.json. Okay. And inside this JSON file, we will be keeping all the specific information uh related to all the agents that are available. Okay. So now let's take this example. So here inside this particular code, you'll be seeing that there are multiple agents. One is Creo AI. First of all, let me close this. Okay. So one one one agent is basically Creo AI, the other agent is Google ADK and the third agent over here is Langraph. So with respect to langraph this is my agent and this agent is basically uh you know if you probably just go ahead and see this agent it converts USD to euro. Okay that is what this agent basically does as a task. It is calling an API over here and internally now see this agent internally can call any kind of MCP things. Okay even we can write the MCP code over here. Okay this is so amazing right? So based on this agent, this agent functionality is to convert USD to euro and you can run this as a specific agent and this agent you can communicate with your uh uh client agent itself. Right? So these are my agents and here we have specifically created our client agent and this client agent has one file which is called as card resolver and it is first of all checking this agent.json file. Now inside this agent dojson file you will be seeing that there will be an information for langraph once we run this particular agent. Okay. So just to show you I'm inside this Python and then I will just go ahead and run this agent slang graph. Along with this I'm also running uh running something called as uh my CLI client agent. Okay. So here you'll be able to see that I'm running this CLI which is my client agent. The client agent is over here CLI host CLI. Okay. So this is my client agent which I'm specifically running. Now this client agent will communicate with this langraph agent. Okay. and we'll see that how things will come up. Okay. So now we will just go ahead and run uv run agent/lang graph. Once I execute this you will be able to see that I'm running this in this particular address. So let me just go ahead and execute this. Okay. And open my browser and here I will just go ahead and write dotwell/ agent JSON. So once I write this here you can see that if I click on pretty print I'm getting all the agent information and how this client agent will be able to discover all the specific agents it is because of this JSON file this configuration file which is having all the necessary information about all the agents that is available currently we are running langraph agent so that is the information that is available over here currency agent helps with exchange rate for currencies all these URL you know ids convert currency currency uh currency exchange rate tools help with exchange values between various currencies, what is the exchange rate between USD and GBP, all the information is basically there. It has all the capabilities like streaming is equal to true, push notification is equal to true and all right now what happens uh once we specifically run this code now see this okay once we are running this code right once we are running the CLI it knows hey what do you want to send to the agent so I'll go ahead and ask a question hey uh please please please please please convert convert convert USD 100 to EUR euro okay and I'll execute this now see Just answer what it will come. It is showing that hey it is usually communicating with this and here you can say I can only provide the exchange rate between two currencies. Can you please uh specify the exact amount? How do you like to convert USD to euro? Okay. So I think it is not able to understand my question but it's okay. We'll go ahead and write uh convert US US to euro. Now I think it should be able to give my answer. So uh let's see. So here you are able to get all the information with respect to exchange rate. Here you can see 1 USD is equal equal equal to88308 and you're able to get the answer. Right? And this is how the communication basically happens. Right? So I hope you like this particular video. We have spoken about so many things. We spoke about how does A2A communication basically works. A2A protocol works. We also spoke about agent discovery with the help of that agent.json file, right? Uh where we specify all the configuration related to the agents. Once that is available, then we also understood how A2A works. We also understood the differences between MCP versus A2A. Just to give you an idea, MCP will be available within a specific AI assistant. You can actually communicate it because here we are accessing some kind of service providers or tools. But agents altogether can be considered combining this MCP with the AI assistant and this will basically communicate with the other agents itself. So yeah, this was it from my side. I hope you like this particular video. This was all about agentto agent protocol. Now coming to the future thoughts. Now see guys, don't worry about so many developments happening in this specific field. These are still evolving. MCP is still evolving. A2A is evolving. And one of the most important thing with respect to MCP and A2A is that even you are a service providers, you need to probably go ahead and implement all your solution with the help of this particular protocol. Similarly, if we are talking about A2A protocol, we have to make sure that all the companies whom we are probably going to access or let's say the collaborator, the aggregator, if they really want to use any specific services between between different companies agents, they need to probably first of all adhere to this particular protocol and probably make their agent compatible to this. Okay, unless and until that is not done, we should we we will not be able to use it, right? So these kind of topics are still evolving. Lot of development will definitely happen. But uh we'll be seeing in the future whatever updates will be coming. I promise you all that I will be making sure to update everything in my YouTube channel. So yeah, this was it from my side. I hope you like this particular video. I will see you in the next video. Thank you. Picker.

---

{
  "summary": [
    "The video introduces Google's Agent2Agent (A2A) protocol, designed for AI agents to communicate and collaborate.",
    "A2A enables secure information exchange and coordinated actions between agents across different platforms.",
    "The video contrasts A2A with the Model Context Protocol (MCP), explaining that MCP allows an agent to access external services, while A2A facilitates agent-to-agent communication.",
    "Agent discovery, facilitated by configuration files (agent.json), allows agents to identify and interact with other available agents.",
    "The presenter provides a coding demo showcasing how agents can interact using A2A, including an example of currency conversion.",
    "Client agents initiate communication and manage tasks, while remote agents provide specific functionalities or access to data.",
    "Both A2A and MCP are evolving protocols crucial for the development of agentic AI applications, requiring adherence from service providers and collaborators."
  ],
  "topics": [
    "Agent2Agent (A2A) Protocol",
    "Model Context Protocol (MCP)",
    "Agent Communication and Collaboration",
    "Agent Discovery",
    "Client and Remote Agents"
  ],
  "formatted_transcript": "Hello guys. So in this video we are going to discuss about this new protocol which is called as agentto aagent protocol which has been launched by Google. We also say it as A2A protocol and uh if you remember like uh recently uh anthropic has also come up with this model context protocol itself. Uh so obviously Google will not be behind you know they'll also come up with something. So that is what they have actually come up with agent to agent protocol. And uh in this specific video we are going to talk about this. We're going to understand how it specifically works. And I'll also be talking about the agenda of this particular video. Right over here you'll be able to see this particular diagram. Uh here there are agents multiple agents and all the agents are basically connected. So obviously you can understand what A2A protocol basically does. It it actually helps you to perform some kind of communication between agents to agents to perform some kind of task. Okay. So going to the agenda what all things we are basically going to discuss in this specific video. So first of all we'll start what is A2A protocol agent to agent protocol how does A2A protocol work difference between A2A versus MCP model context protocol because we also need to understand this you know uh like how this two specific A2A protocol and MCP protocol differ and then we will also be seeing some kind of demo with coding where we'll run this kind of we'll see how this A2A agent communication will basically happen and then I'll also talk about the future thoughts uh with respect to A2A protocol call and MCP protocol. So step by step we will be seeing all the specific points. It is important for you all to understand because this protocol will be really necessary as we go ahead and since this agentic AI application completely is evolving this will be very much important both MCP and A2A. Okay. So it is necessary that you please watch this video till the end. Okay. So first of all uh let's go back to the docu documentation here. You can I'll just read some of the lines and then with basic examples I'll try to showcase each and everything. Okay. So AI agents offers a unique opportunity to help people to be more productive by autonomously handling many daily recurring or complex tasks. This we already know right? If you're building some agentic AI applications there you'll be able to see that AI agents will be able to perform many many tasks. Even AI agents will also be able to communicate with the other AI agents to solve some complex task itself. Okay. Now they are launching a open protocol which is called as agent to agent with support and contribution from more than 50 technological partners. So Google when they came up with this thing um they have also a lot of technological partners that is good you know like bigger companies like Atlaxian, Box, Coair, Intude, Langchain, MongoDB, PayPal, Salesforce, SAP, Service Now, UK uh UKG and Workday right and then lot of other companies who and this A2A protocol will allow AI agents to communicate with each other securely exchange information coordinate actions on top of various enterprise platform or applications. So what this is basically doing is that this protocol will actually help one AI agent to communicate with the other AI agents. Now just imagine if some of the functionalities from this particular company is implemented in a form of agents and that is exposed to the other agents then the other agent will be able to collaborate and perform various task itself. Again this is just a thought process but I will show you multiple examples. I will probably draw each and everything for this. Okay. Here also it is written A2A is an open protocol that complements anthropics model context protocol. As I said that it is just like complementing on top of MCP. Okay. So we'll also understand the difference. Now let me go back to my uh coding screen and over here you'll be able to see writing screen. Sorry here I have probably given the definition. It is an open protocol that complements anthropic model context protocol which provides helpful tools and context to agents. Drawing on Google internal expertise in scaling agentic system. We designed the A2A protocol to address the challenge we identified in deploying last scale multi-ere system for our uh customers. Now to understand what is A2A protocol. Okay. So let's say that I have this as one of my agent and this is my another agent. Okay. Let's say right now u if this is my agent A. Okay. So I will just go ahead and write this is my agent A. Okay. So let's say this is my agent A and this is my another agent that is B. Okay. Let's say this specific company you know is just like a booking website. Okay. Booking website. One of the booking website that we commonly use in India. Let's say as an example it is make my trip. Okay, it can be make my trip. It can be a it can be various other sites you know it is up to us like whatever site you specifically refer. Now let's say that the input given by the user is that hey please travel my plan or plan my travel. Sorry I'm not saying travel my plan my travel. Okay. So, plan my travel. Okay. To some countries. Let's say I want to probably go to Europe. Okay. For 7 days. And here we will say that one of the important goal is with minimum cost. Okay. With minimum cost. So I tell this specific agent, this agent is of this particular booking website and I'm telling them to perform this specific task. Okay. Now what agent A will basically do, it will go ahead and communicate, right? It will probably go ahead and communicate with different types of agents. This is my agent A B. There may be other agents over here. There may be one more agent somewhere here. Right? and this agents just imagine I'll go ahead and write this is my agent B then let's say this is this is my agent other agent over here okay please try to understand this okay so this is my agent uh C this is my agent D this is my agent E okay okay now with respect to this particular plan travel first of all we need to see that what are the intermediate steps okay first of all we think that we need to book flights right so obviously I need to probably go ahead and communicate with an agent which is responsible in booking flight right so here I will be having this specific agent and this agent will be developed by a flight booking company let's say that this is nothing but Emirates okay let's say this is Emirates Emirates okay Emirates is one flight booking industry right so this agent will first of all try to find out that hey whom do I need to probably communicate with for booking flights? Okay. And let's say this agent A will be able to identify there are two different agents C and E which will be which are nothing but these are agents of some other companies. Okay. Some other booking flight companies. So here is Emirates. Let's say this is another flight which is like Air India. Okay. Air India. So Emirates and Air India will be able to probably tell give some kind of information with respect to the price. Okay, with respect to the price. So let's say that over here I am getting some kind of information from Emirates. Hey, this is what is the price that I will be getting since uh these are the agents that is probably communicating over here. And remember this Emirates will have internal data information, right? And this Emirates, this agent that is created, it will be specifically for this Emirates. This agent E will be specifically an agent for Air India, right? And this is how this agent is able to identify it is able to take the information from both of these particular agents and then it'll get a kind of response over here and this response will be having whichever will be the minimal cost it will take that specific options and obviously from Emirates and Air India I feel Air India will give us a affordable cost with respect to the flight travels. Okay. Now the second thing is that we need to book we need to book hotels. Okay, we need to book hotels. Now for this hotels let's say that I have B and D agents available right and how it'll identify which all agents are available that we specifically called as agent discovery that also I will talk about it. Okay. So through agent discovery how this agent discovery is done I will also talk about it. But let's say that from this agent discovery. So here we basically go ahead and discover agents and we will get to know that how many different types of agents are there. Now let's say based on this agent discovery we know that for booking FL hotels there are two different kind of agents available. Okay let's say this is one this is the other one. Okay let's say this agent is specifically belonging to Taj hotels. Okay I'm just taking some example and this agent is basically belonging to let's say Marriott. Okay, variate. Now if these two agents are available, so what this agent will do is that it'll try to communicate with them and based on this this will try to get some kind of information and whichever is the affordable one, it will select those. Right? So this is how agentto agent communication will specifically happen. So now similarly you can probably consider any kind of different kind of use case uh considering a aggregator platform you know. So let's say make my trip is one kind of aggregator platform which will actually help you to connect to flights, hotels, um you know even probably plan the entire trip in short right similarly you can take another example like let's say for jobs you know for jobs in India we definitely have lot of job portals right so there are job portals now this job portals can basically create an agent and connect to various companies agent which are responsible in hiring things you know hiring hiring candidates candidates right and based on that right let's say if Dell if uh there's a company like IBM there's a company like uh HCL there's a company like Sapient if they want a specific requirement what they can do is that that specific agent can probably communicate with this job portal agent and probably provide uh candidates with respect to that right uh similarly let's say that over here I have this particular job portal again the communication will happen in that same way right agentto agent communication that what exactly A2A basically means. Okay. Now if I go further right how A2A works I've already spoken about it you know by drawing all the specific diagrams but if I talk about architecture point of view there are two types of agent one is the remote agent and once one is a client agent. So this agent what we are speaking speaking about is it is basically called as client agent. Okay. So this is nothing but it is it is your client agent. Client agent. Okay. And these all agents that you'll be seeing it is basically nothing but it is a remote agent. Okay. This is nothing but remote agent. Right. How client agent is able to see so many agents? It is from this agent discovery mode. Okay. Through the agent discovery, this client will be able to understand how many different agents are available. Right? I will talk more about this agent discovery. how this agent discovery is you know there is some configuration files we need to probably set it up for this okay every agent needs to probably set a configuration file for this and we'll talk about that particular file and this file will be exposed to all the other agents now this agent is communicating agents now this agent is communicating to this right this agent can also communicate to this agent this agent can also communicate to this agent this agent can also communicate to this agent right this agent can also communicate to this agent right this agent C to D also can communicate C to E also can communicate so how every agent will know that there is some other agent because of this configuration file uh settings you know and we'll talk about that particular configurations file once we probably do the demo in in respect to coding right now let's understand this so there is a client agent there is a remote agent uh these all are specifically remote agent the client agent has some more functionalities you know all the communication that basically happens between the client agent and remote agent will have the secure collaboration it'll make sure to manage the task and state management user uh experience negotiation And even capability discovery as I said capability discovery is like the client agent will know that how many remote agents are available and this is through some kind of configuration configuration. Okay. And uh this configuration uh will be nothing but it is something like it will be exposing the entire agent card. Okay. We basically say it as agent card. I will talk about it as we go ahead. Okay. So this gives a clear idea about how A2A specifically works. But now one question that arises what is the difference between M2 MCP versus A2A right? Now let's say that I have this agents right so I have this agent one I have this agent two right I have this agent three right let's say these all are my agents now within this particular agent let's say uh let's say this is my agent A B C right and with respect to MCP also let's say I will go ahead and select a circle now agent to agent communication is definitely happening using a A2A protocol but when we talk about MC MCP MCP basically means what? Let's say this specific agent requires some of the services that needs to be required from a service provider. Let's say this is my service A, service B, this is service C, right? And let's say this agent wants to specifically use this service. Okay. Oops. Just a second. Okay. So this is my agent A and let's say this agent A wants to access uh is connected to various services and uh based on the services it requires some kind of information let's say this is internet search let's say this particular service is about it's uh DB connection local DB connection it can be third party APIs anything as such right so this agent A is basically communicating with the specific services with the help of MCP P the model context protocol and obviously if you remember in my previous video I've also discussed about MCP right and similarly this is my agent B okay agent B now inside this agent B let's say that I have different different different uh uh a uh services that I really want to use over here also so let's say this is my service A service B and this is my service C okay so different different agents uh will basically be connected over here again with the help of MCP and this will basically be doing this kind of communication and you know that what is MCP? MCP is nothing but model context protocol. It is a protocol which is specifically used by an agent or an AI assistant. I can also say AI assistant to probably get some kind of context from this external tool providers or service providers, right? So if I probably consider this this can entirely become okay one agent. So let's go ahead and draw this diagram also so that you'll be able to understand it very much clearly. So this can be my entire one kind of agent and internally an agent can have many MCP connections with respect to different MCP servers. Right? So this internally agent can have different different connections to the service provider right and this will happen internally to this particular agent all the specific communication will happen when we are talking about agentto agent communication that basically means this agent is communicating to the other agent right but it will not have any idea like what all MCP servers this will basically be connected to we are just connecting this agent to agent communication to get the information or work done by this specific specific agent right similarly for this specific agent also we are doing in that similar way right but internally this agents man can have any number of MCP connections okay so just to show you one very good diagram so let's say that this is one of my agent let's say this is one of my agent let's say this is one of my agent I can I can probably change this specific diagram to something like this let's say here I have one I have one MC uh one MCP server two MCP server and this is my AI assistant or agent itself Okay, this is my AI assistant or agent that is communicating to this with the help of MCP. Okay. And this is entirely my agent. One agent. Let's say this is agent A. This agent A is communicating to agent B to get the information. But internally this may be connected to some kind of MCPS also. Right. So MCP and A2A. I hope you have understood the basic difference. Okay. Now the question rises Kish can you show some example? And we definitely need to see some kind of demo with coding. Right. So for this what I've done is that I've directly taken the entire repository that has been given by Google. Okay. The first thing is that we'll try to understand how this discovery will basically happen. How this agent discovery will basically happen. I told you that we are going to set up some configuration file and even what Google has basically done is that whenever you create any kind of uh agents there will be one JSON file on this name wellknown agent.json. Okay. And inside this JSON file, we will be keeping all the specific information uh related to all the agents that are available. Okay. So now let's take this example. So here inside this particular code, you'll be seeing that there are multiple agents. One is Creo AI. First of all, let me close this. Okay. So one one one agent is basically Creo AI, the other agent is Google ADK and the third agent over here is Langraph. So with respect to langraph this is my agent and this agent is basically uh you know if you probably just go ahead and see this agent it converts USD to euro. Okay that is what this agent basically does as a task. It is calling an API over here and internally now see this agent internally can call any kind of MCP things. Okay even we can write the MCP code over here. Okay this is so amazing right? So based on this agent, this agent functionality is to convert USD to euro and you can run this as a specific agent and this agent you can communicate with your uh uh client agent itself. Right? So these are my agents and here we have specifically created our client agent and this client agent has one file which is called as card resolver and it is first of all checking this agent.json file. Now inside this agent dojson file you will be seeing that there will be an information for langraph once we run this particular agent. Okay. So just to show you I'm inside this Python and then I will just go ahead and run this agent slang graph. Along with this I'm also running uh running something called as uh my CLI client agent. Okay. So here you'll be able to see that I'm running this CLI which is my client agent. The client agent is over here CLI host CLI. Okay. So this is my client agent which I'm specifically running. Now this client agent will communicate with this langraph agent. Okay. and we'll see that how things will come up. Okay. So now we will just go ahead and run uv run agent/lang graph. Once I execute this you will be able to see that I'm running this in this particular address. So let me just go ahead and execute this. Okay. And open my browser and here I will just go ahead and write dotwell/ agent JSON. So once I write this here you can see that if I click on pretty print I'm getting all the agent information and how this client agent will be able to discover all the specific agents it is because of this JSON file this configuration file which is having all the necessary information about all the agents that is available currently we are running langraph agent so that is the information that is available over here currency agent helps with exchange rate for currencies all these URL you know ids convert currency currency uh currency exchange rate tools help with exchange values between various currencies, what is the exchange rate between USD and GBP, all the information is basically there. It has all the capabilities like streaming is equal to true, push notification is equal to true and all right now what happens uh once we specifically run this code now see this okay once we are running this code right once we are running the CLI it knows hey what do you want to send to the agent so I'll go ahead and ask a question hey uh please please please please please convert convert convert USD 100 to EUR euro okay and I'll execute this now see Just answer what it will come. It is showing that hey it is usually communicating with this and here you can say I can only provide the exchange rate between two currencies. Can you please uh specify the exact amount? How do you like to convert USD to euro? Okay. So I think it is not able to understand my question but it's okay. We'll go ahead and write uh convert US US to euro. Now I think it should be able to give my answer. So uh let's see. So here you are able to get all the information with respect to exchange rate. Here you can see 1 USD is equal equal equal to88308 and you're able to get the answer. Right? And this is how the communication basically happens. Right? So I hope you like this particular video. We have spoken about so many things. We spoke about how does A2A communication basically works. A2A protocol works. We also spoke about agent discovery with the help of that agent.json file, right? Uh where we specify all the configuration related to the agents. Once that is available, then we also understood how A2A works. We also understood the differences between MCP versus A2A. Just to give you an idea, MCP will be available within a specific AI assistant. You can actually communicate it because here we are accessing some kind of service providers or tools. But agents altogether can be considered combining this MCP with the AI assistant and this will basically communicate with the other agents itself. So yeah, this was it from my side. I hope you like this particular video. This was all about agentto agent protocol. Now coming to the future thoughts. Now see guys, don't worry about so many developments happening in this specific field. These are still evolving. MCP is still evolving. A2A is evolving. And one of the most important thing with respect to MCP and A2A is that even you are a service providers, you need to probably go ahead and implement all your solution with the help of this particular protocol. Similarly, if we are talking about A2A protocol, we have to make sure that all the companies whom we are probably going to access or let's say the collaborator, the aggregator, if they really want to use any specific services between between different companies agents, they need to probably first of all adhere to this particular protocol and probably make their agent compatible to this. Okay, unless and until that is not done, we should we we will not be able to use it, right? So these kind of topics are still evolving. Lot of development will definitely happen. But uh we'll be seeing in the future whatever updates will be coming. I promise you all that I will be making sure to update everything in my YouTube channel. So yeah, this was it from my side. I hope you like this particular video. I will see you in the next video. Thank you. Picker.",
  "notable_quotes": [
    "Obviously you can understand what A2A protocol basically does. It it actually helps you to perform some kind of communication between agents to agents to perform some kind of task.",
    "A2A is an open protocol that complements anthropics model context protocol.",
    "MCP is nothing but model context protocol. It is a protocol which is specifically used by an agent or an AI assistant to probably get some kind of context from this external tool providers or service providers",
    "MCP will be available within a specific AI assistant... But agents altogether can be considered combining this MCP with the AI assistant and this will basically communicate with the other agents itself.",
    "Now see guys, don't worry about so many developments happening in this specific field. These are still evolving. MCP is still evolving. A2A is evolving."
  ]
}