Setting the file. One moment.
Subchapter 53.38
foundry-agent/create/references/tools/managed-harness-agent/tool-a2a.mdMarkdown1 KBView on GitHub
tools:
- type: a2a_preview
name: specialist
description: Handles specialist requests.
base_url: https://<peer-endpoint>
agent_card_path: /.well-known/agent-card.json
project_connection_id: <optional-connection>Same-project or public peers may not need a connection. For an authenticated peer, use a connection supplied by the user or already referenced by the Agent.
Create a remote A2A connection only when explicitly requested:
azd ai connection create <connection-name> \
--kind remote-a2a \
--target "<peer-a2a-endpoint>" \
--auth-type <user-entra-token-or-agentic-identity> \
--audience "https://ai.azure.com" \
--metadata "ApiType=Azure" \
--metadata "type=custom_A2A" \
--metadata "AgentCardPath=/agentCard/v1.0" \
--project-endpoint "<project-endpoint>"Grant the invoking user or Agent identity the Foundry Agent Consumer role on the peer project.
The target must already expose an A2A endpoint and Agent Card. For a Foundry peer that does not, follow Enable Incoming A2A.