Azure AI Foundry Agents

The Azure AI Foundry Agent Service calls an outside capability through an OpenAPI 3.0 tool, and interoperates with other agents over A2A. This platform generates both from the registry that the gateway checks at call time.

#Option A - an OpenAPI tool

curl -H "authorization: Bearer $TOKEN" https://app.authprior.com/api/orchestrate > priorauth.json

1. Add the tool
   Register priorauth.json as an OpenAPI tool on your Foundry agent. Each registered agent becomes one operation, described as an outcome.
2. Carry the credential on a connection
   Attach the channel token through a Foundry connection backed by Key Vault, or a managed identity - never in the agent's instructions. Foundry authenticates as a channel principal bound to specific practices.

#Option B - A2A

Register this platform as a remote A2A agent from its card at /api/agent-card; the service drives the task lifecycle and surfaces input-required back to your workflow.

#Two things that are not optional

1. Treat 202 as "poll"
   An operation held at the release gate returns 202 with a task id. The tool must poll tasks/get; the default retry-on-non-200 behaviour creates duplicate tasks against a practice that believed it asked once.
2. The gate is not a setting
   There is no release operation to call. input-required clears only when a licensed human at the practice releases that exact submission - the caller, human or agent, cannot satisfy it.

#What Foundry should not be asked to do

• Approve or release anything.
• Enumerate practices.
• Poll faster than the payer moves.