[ AI tools · Frameworks and platforms ]
Agents From Different Teams, Working as One
The Agent-to-Agent protocol lets agents built on different frameworks, by different teams or companies, find each other, hand over tasks and report progress. We used it in NextLM and build A2A agents for clients.
Retailer
Order agent
Skills
- track_order
- request_return
/.well-known/agent-card.json
Logistics partner
Courier agent
Skills
- book_pickup
- reschedule
/.well-known/agent-card.json
Task: book a return pickup for order 48213
submittedIllustrative exchange between two companies’ agents.
A Common Language for Agents to Delegate Work
Agent2Agent, started by Google and now hosted by the Linux Foundation, lets agents built on different frameworks find each other and hand over tasks. It starts with an agent card; point at a line to read it.
{}
Agent cards
A public description of what an agent can do, how to reach it and how to authenticate.
Simplified example agent card.
How One Agent Hands a Job to Another
- 01
Discover
The order agent reads the courier agent's card and sees it can book pickups.
- 02
Send the task
It sends a message with the order, address and preferred dates.
- 03
Work in progress
The courier agent checks slots and streams status updates back.
- 04
Ask for input
Two slots fit. The courier agent asks which one the customer prefers.
- 05
Complete
The order agent replies, the pickup is booked and a label comes back as an artifact.
a2a message log
- GET /.well-known/agent-card.json
- message/send { "task": "book_pickup" }
- status: "working" · checking slots
- status: "input-required"
- status: "completed" · artifact: label.pdf
Two Protocols, Two Different Jobs
A2A
Agent to agent
For delegating whole tasks to another agent that has its own reasoning, tools and owner, possibly in another company.
“Courier agent, book this return.”
MCP
Agent to tool
For giving one agent direct access to tools and data, such as a database lookup or an API call, with no reasoning on the other side.
“Call get_order(48213).”
Most production systems use both: A2A between agents, MCP for each agent’s own tools.
A2A in a Product We Built
NextLM, the AI prospecting product we built, uses A2A so its agents can hand tasks to one another across services, with MCP for each agent’s own tools.
Read the NextLM case study- 01
Partner integrations
Your agent books, checks or orders through a supplier's agent instead of a brittle API integration.
- 02
Cross-team agents
Sales, support and finance agents built by different teams that still cooperate.
- 03
Mixed frameworks
An ADK agent, a LangGraph agent and a Copilot Studio agent working on one request.
- 04
Marketplaces
Platforms where third-party agents offer skills to your users.
[ Agent card first ]
Every Agent Introduces Itself Before It Works
We start every A2A project by writing the agent cards: what each agent can do, how it authenticates and what it will never do.
Design Your Agent CardsWiring Agents Together, Alone or With Us
- Agent cardsAloneWritten once, drift from what the agent really doesWith usGenerated from the agent's code and checked in CI
- Auth between agentsAloneShared API keys passed aroundWith usOAuth per caller, scoped to the skills it may use
- Long tasksAloneRequests that time out after a minuteWith usTask states with streaming and push updates
- When something failsAloneNobody knows which agent dropped itWith usEvery task traced across agents, end to end
Packages from $1,500; custom projects from $8,000.
A2A Questions
What is the A2A protocol?
Agent-to-Agent is an open protocol for agents to discover each other, delegate tasks and exchange results, regardless of which framework or company built them. It was started by Google and is now hosted by the Linux Foundation.
How is A2A different from MCP?
MCP connects an agent to tools and data. A2A connects an agent to another agent that has its own reasoning and tools. They complement each other and are often used together.
Which frameworks support A2A?
Google's ADK, Microsoft Agent Framework, LangGraph, CrewAI and others support A2A, along with major cloud agent platforms.
Is it safe to let a partner's agent act for us?
With the right controls: each agent card declares its auth scheme, calls use standard enterprise authentication, tasks have scoped permissions and every exchange is logged on both sides.
Have you used A2A in production?
Yes. NextLM, the AI prospecting product we built, uses A2A so its agents can hand tasks to one another across services.
Do we need A2A for a single agent?
No. A single agent with a few tools only needs MCP or direct function calling. A2A pays off when separate agents, often with separate owners, must cooperate.
Let's build your next product together
Book a free strategy call and leave with a clear plan and estimate. No commitment.
- 01Pick a time that suits you
- 0230 minutes on scope, stack, timeline and budget
- 03Fixed-price proposal, NDA on request