Connect to another agent
With the A2A Agent component a NappAI flow talks to an A2A agent: it sends it a message and gets its answer. It can be another flow of yours published as an agent, one from another NappAI account, or an agent on another platform.
1. Add A2A Agent
| Field | What to write |
|---|---|
| Agent URL (required) | The address of the agent or of its card. NappAI reads the card and knows how to talk to it. |
| Message | What you ask it. If the component is an agent’s tool, that agent writes it. |
| Keep Conversation | Keeps the same conversation with the remote agent across the messages of a session, so it remembers what came before. On by default. |
| Max Wait (seconds) | The longest to wait for the remote agent to finish. If it does not finish in time, it is asked to stop and the component reports the error. |
The credential
If the remote agent asks for a key, create an A2A Agent credential with it in the API key or token field. NappAI sends it the way the agent’s card says. Public agents do not need it.
To talk to an agent published in NappAI, the credential holds an API key of the account that owns that agent.
2. Use it as an agent’s tool
The usual way is to connect its Tool output to an agent’s Tools input. The agent reads the remote agent’s description and decides when to hand it something, with what message, and what to do with the answer.
Tell it in its instructions when to delegate. For example: “To prepare an itinerary, always ask the travel planner with your tool”.
To build a team with agents from several platforms, add one A2A Agent component per remote agent and connect them all to the same agent.
Also as a step of the flow
Without connecting it to an agent, A2A Agent runs as one more step: it sends the Message and its Response output gives the remote agent’s answer. The Agent Card output gives the agent’s card, in case you want to show or keep its details.
What the remote agent can answer
- The answer, with the text and, if it sends them inside the message, its attached files.
- That it needs more information. The answer says so and what is missing. If the component is a tool, the agent can ask the user and call it again.
- That it cannot do it or that it failed, with its reason.