Skip to content

Publish an A2A agent

Any flow that holds a conversation, with a Chat Input that receives the message and a Chat Output that gives the answer, can be published as an A2A agent. Other agents will hand it tasks as if they were one more user of the chat.

1. Add A2A Server

Add the A2A Server component to the flow. It is not connected to anything: its job is to describe the agent.

FieldWhat to write
Agent NameThe name other agents see. Empty, the flow’s name is used.
Description (required)What the agent does and when it is worth asking it something. Other agents read it to decide whether to hand it a task.
SkillsIts skills, one per line, as Name: what it does.
Example RequestsExample requests, one per line, to show how to ask it things.
TagsWords, separated by commas, that help others find it.
VersionThe version shown on its card.

2. Deploy a version

A2A calls always run the deployed version of the flow. Deploy a version from the right bar of the editor and, if you later change the flow or the agent’s description, deploy another one. See Versions and publishing.

3. Share its address

  1. Press API Access, leave the consumption mode on Production and open the A2A Server tab.
  2. Choose an API Key. See API keys.

The tab gives you:

  • Agent card (public): the address of its card. Anyone can read it, and with it any compatible platform knows what your agent does and how to talk to it.
  • Agent URL: the address tasks are sent to.
  • Examples to call it from other platforms, with the chosen key already included.

Handing it tasks needs an API key of your account. From another NappAI flow it is used with the A2A Agent component: see Connect to another agent.

How it handles tasks

  • Each of the caller’s conversations has its own memory: if the flow has memory, it remembers what was said earlier in that conversation.
  • The caller can wait for the answer or collect it later by checking the task: long tasks are not cut off.
  • Files can be sent along with the message. They reach the flow through the Binary output of the Chat Input.
  • Each task appears in Executions, like any other run.

NappAI agents work with the main A2A implementations, such as Google’s and the protocol’s official kit.