Ngiyazi i-Google i-open source agent ye-agent protocol, ebizwa ngokuthi
Ngingathanda kakhulu, ngakho-ke ngifuna ukuyifaka ngakuvamile futhi ukuhlola ukuthi isikhundla se-A2A ku-universe ye-agent. Ngakho-ke siphindeza ukuthi i-A2A isebenza kanjani emzimbeni emzimbeni emzimbeni, kanye nokuxhumana ne-MCP.
Ukuhlehlela A2A
Ukusetshenziswa kwe-A2A kulinganayo ne-MCP. Ungasebenza ama-A2A ama-agents/i-server amancane, bese i-A2A client ingasebenzisa kubo bonke. Izindaba ezinhle zihlanganisa ukuthi ngokuvamile unemibuzo yokusebenza kwe-A2A ama-agents kanye ne-A2A client.
Ukusebenza kwe-A2A Agents (i-server)
Ngithunyelwe zonke amabhizinisi amabhizinisi amabhizinisi amabhizinisi
- ikhaya
- Umphathi we-Google ADK angakwazi ukunikela imibuzo yakho ikhaya
- I-CrewAI agent okuyinto inokufunda ukukhiqiza isithombe ikhaya
- I-Agent ye-LangGraph enokufunda i-ultimate exchange rate ikhaya
Umthengi we-A2A inikeza emhlabeni ukuthi izinzuzo zayo zithunyelwe nge-"Agent Card" ku-JSON format. Njengesibonelo, ikhadi le-agent ye-google ADK ibona lokhu:
{
"name": "Reimbursement Agent",
"description": "This agent handles the reimbursement process for the employees given the amount and purpose of the reimbursement.",
"url": "http://localhost:10002/",
"version": "1.0.0",
"capabilities": {
"streaming": true,
"pushNotifications": false,
"stateTransitionHistory": false
},
"defaultInputModes": [
"text",
"text/plain"
],
"defaultOutputModes": [
"text",
"text/plain"
],
"skills": [
{
"id": "process_reimbursement",
"name": "Process Reimbursement Tool",
"description": "Helps with the reimbursement process for users given the amount and purpose of the reimbursement.",
"tags": [
"reimbursement"
],
"examples": [
"Can you reimburse me $20 for my lunch with the clients?"
]
}
]
}
Ukushaja i-A2A Client demo app
Thola ku-client. Imibuzo yokwenza i-web app ye-demo kusebenza lapha.
Uma ifakiwe i-web app, ungakwazi ukufinyelela kwalo kusuka ku-browser yakho. I-client ibona i-Gemini AI Studio nge-Google Material design.
URL: localhost:12000
Okokuqala, kuqala, kufanele ukongeza zonke ama-agents kumakhasimende ngokuvumelana ne-URL yayo ye-base. Njengoba ku-my case ngitholela zonke ama-agents e-locally, ama-URL yayo yayo yayo yayo:
- ikhaya
- I-Google ADK ye-localhost:10002 ikhaya
- umphathi wedivayisi:10001 ikhaya
- LangGraph Localhost:10000 ikhaya
I-Side Note: ngaphakathi kwe-protocol, i-URL ye-ultimate ibonakala nangokunye:
https://localhost:10002/.well-known/agent.json
Ngaphezu kwalokho, ungakwazi ukubonaall three agentsKuyinto ezihambelana:
Uyakwazi ukubonachat historyNgiya
Zonke ievent list
Zonke itask list
SettingsUkulungele Basic
Ukubuyekezwa kwe-Google ADK Agents for Expenses Claim
Ukubuyekezwa kwe-LangGraph ye-Forex
Ukuhlolwa kwe-CrewAI ye-agent ye-image generation
I-combo test ye-multi-agent
Ngingathanda ukuthi i-client ye-A2A iyakwaziuse multiple agents to achieve a single goalNgakho ngithole ukuthi kungenziwa ukuxhumana ne-expense claim agent ne-forex rate agent. Futhi kungenziwa ukusebenza.
Ukusebenza kwami »Ukubuyekezwa kwe-cost for a beer in Germany while on a business trip, 5 euro, April 4 2025”. Ukusabela wahlanganyela amahora embalwa, futhi ekugcineni kwenza inani elilodwa le-US dollar ku-expense claim form.
Ukubuyekezwa kwe-A2A
Ngithanda ukuthi i-A2A iyimodeli ye-Client-Server efanelekileyo enokuthuthaza kanye nokuhambisana ngokushesha. I-Client ayinempumelelo ukucacisa nokulanda ama-agents / ama-server.
Isakhiwo se-agent kuyinto elula kakhulu ngokuvumelana ne-URL yokufaka, futhi i- "Agent Card" ivimbela ukuxhumana kwe-context. Futhi ungakwazi ukwengeza futhi ukunciphisa ama-agents ngemuva kwe-client iyasungulwa.
Ngokusho kwe-demo yamanje, kubaluleke kanjani ama-agents ukuxhumana nabanye kanye nokufinyelela izicelo ezinzima. I-client ikhulumeza ngamunye ama-agents ngokulinganayo izicelo ezahlukene, ngakho-ke ngokulinganiswa ne-multi-tool calling.
Ukuqhathanisa A2A nge-MCP
Ngaphezu kwalokho, i-A2A iyahambisana ne-MCP, okuthiwa i-A2A.
Uma i-A2A kanye ne-MCP zihlanganisa ukuthuthukiswa kwamakhemikhali we-AI, ngokwemvelo zihlanganisa nezidingo ezahlukile. I-A2A isebenza ngezinga le-agent-to-agent, ngokuvimbela ukuxhumana phakathi kwamakhemikhali ezingenalutho, lapho i-MCP isebenza ngezinga le-LLM, ngokuvimbela ukuthuthukiswa kwe-context ne-capacities ye-languages eyodwa.
Futhi ukunikezela imibuzo yayo eziyinhloko kanye nemibuzo ngokuvumelana nemiprotokoloni yayo:
ikhayaIzici A2A MCP Ukusetshenziswa Okokuqala I-Case-Agent-to-Agent Ukuxhumana nokuxhumana nokuxhumana Ukunikeza isixhobo kanye nezixhobo (i-API/SDK) ku-LLMs Core Architecture Client-server (i-agent-to-agent) Client-host-server (i-application-LLM-external resource) I-Standard Interface JSON Ukucaciswa, I-Agent Card, I-Task, I-Messages, I-Artifacts JSON-RPC 2.0, Izici, I-Tools, I-Memory, I-Prompts Izici Multimodal, Dynamic, Security Collaboration, Task Management, Capacity Discovery Modularity, Izinzuzo zokhuseleko, UkuxhumanaUkuhlobisa
A2AA2A
ikhayaUkubuyekezwa
Ukubuyekezwa
Ukusetshenziswa Okokuqala Case-Agent-to-Agent Ukuxhumana nokuxhumana Ukunikeza umklamo kanye nezixhobo (i-API / SDK) ku-LLMsUkusetshenziswa okokuqala CaseUkusetshenziswa okokuqala Case
I-agent-to-agent Ukuxhumana nokuxhumanaI-agent-to-agent Ukuxhumana nokuxhumana
ikhayaUkunikeza umklamo nezindlela (i-API / SDK) ze-LLMs
Ukunikeza umklamo nezindlela (i-API / SDK) ze-LLMs
I-Core Architecture I-Client-server (i-agent-to-agent) I-Client-host-server (i-application-LLM-resource ye-external)I-Core ArchitectureI-Core Architecture
ikhayaI-Client Server (I-Agent ku-Agent)
I-Client Server (I-Agent ku-Agent)
ikhayaI-Client-host-server (i-application-LLM-i-resource ye-external)
I-Client-host-server (i-application-LLM-i-resource ye-external)
I-Standard Interface JSON Specification, I-Agent Card, I-Task, I-Messages, I-Artifacts JSON-RPC 2.0, I-Resources, I-Tools, I-Memory, I-PromptsikhayaI-Interface ye-Standard
I-Interface ye-Standard
ikhayaUkucaciswa kwe-JSON, I-Agent Card, I-Task, I-Messages, I-Artifacts
Ukucaciswa kwe-JSON, I-Agent Card, I-Task, I-Messages, I-Artifacts
I-JSON-RPC 2.0, Izinsiza, Izinsiza, I-Memory, I-PromptsI-JSON-RPC 2.0, Izinsiza, Izinsiza, I-Memory, I-Prompts
Iziqu ze-Multimodal, Dynamic, Secure Collaboration, Task Management, Capability Discovery Modularity, Izinzuzo ze-Security, Reusability of Connectors, SDKs, Tool DiscoveryImikhiqizo KeyImikhiqizo Key
I-multimodal, dynamic, i-secure collaboration, i-task management, i-capability discoveryI-multimodal, dynamic, i-secure collaboration, i-task management, i-capability discovery
I-Modularity, Izinzuzo ze-Security, Ukusebenzisana Kwe-Connectors, I-SDKs, Ukuhlola AmathuluziI-Modularity, Izinzuzo ze-Security, Ukusebenzisana Kwe-Connectors, I-SDKs, Ukuhlola Amathuluzi
ikhayaI-Protocol ye-CommunicationikhayaHTTP, JSON-RPC, SSEikhayaI-JSON-RPC 2.0 over stdio, i-HTTP nge-SSE (noma i-HTTP ye-streaming)ikhayaI-Protocol ye-CommunicationI-Protocol ye-Communication
HTTP, JSON-RPC, SSEHTTP, JSON-RPC, SSE
I-JSON-RPC 2.0 over stdio, i-HTTP nge-SSE (noma i-HTTP ye-streaming)I-JSON-RPC 2.0 over stdio, i-HTTP nge-SSE (noma i-HTTP ye-streaming)
ikhayaUkusebenza FocusikhayaI-Asynchronous Communication ye-Load HandlingikhayaUkulawula okuphakeme kwe-context, ukucubungula okuphakeme, ukucubungula okuphakemeikhayaUkusebenza FocusUkusebenza Focus
I-Asynchronous Communication ye-Load HandlingI-Asynchronous Communication ye-Load Handling
Ukulawula okuphakeme kwe-context, ukucubungula okuphakeme, ukucubungula okuphakemeUkulawula okuphakeme kwe-context, ukucubungula okuphakeme, ukucubungula okuphakeme
I-Adoption & I-Community I-Big Initial Industry Support, i-Ecosystem Emerging Ukuvumelela okuphakeme kwebhizinisi jikelele, i-Community eyenziwe ngokusheshaikhayaI-Adoption & I-Community
I-Adoption & I-Community
Ukubuyekezwa kwama-Industry, i-Ecosystem yokuqalaUkubuyekezwa kwama-Industry, i-Ecosystem yokuqala
Ukusetshenziswa okuphakeme kwebhizinisi lonke, isixazululo esikhulu esikhuluUkusetshenziswa okuphakeme kwebhizinisi lonke, isixazululo esikhulu esikhulu
Imibuzo
Nangona i-Google ibonisa ukuthi i-A2A kuyinto i-protocol ye-MCP ye-cost-free, ukuhlolwa yami yokuqala kubonisa ukuthioverwhelmingly overlapping in purpose and features. Zonke zihlanganisa nezidingo zamabhizinisi zezicelo ze-AI ukusetshenziswa ama-agents eziningana nezixhobo ukufinyelela izici ezinzima. Ngoku-akhawunti, zihlanganisa izindlela ezinhle zokufaka kanye nokufunda ama-agents nezinye nezixhobo ngaphandle kwe-manual configuration.
I-MCP uqala ngokushesha futhi iyathola ukwesekwa kakhulu kumazwe we-developer kanye nama-entrepreneurs. I-A2A kuyinto omdala kakhulu, kodwa iyathola ukwesekwa okusheshayo kumakhasimende amaningi we-Google Cloud Enterprise.
Ngingathanda ukuthi lokhu imibuzo enhle kumadivayisi, ngoba ngeke uthole ukhetho okwengeziweopen and standard agent-agent protocolsImininingwane kuphela ungayifaka ukuthi iyiphi izimpendulo kuyimfuneko, noma kungase ngempumelelo ku-standard eyodwa.
