Hosted Agent to Hosted Agent A2A delegation returns HOSTED_AGENT_NOT_SUPPORTED

Manivannan M 0 Reputation points
2026-09-25T14:04:13.1966667+00:00

We have a multi-agent solution in Azure AI Foundry where an orchestrator and multiple business specialists are independently deployed as kind: hosted agents.

Architecture: Agent A (Hosted)

├── Agent B (Hosted)

├── Agent C (Hosted)

├── Agent D (Hosted)

└── other specialist agents (Hosted)

The specialist agents need to remain independently hosted because they contain custom Python logic, MCP tools, security/RBAC handling, and separate deployment lifecycles.

We followed Microsoft's current GA A2A v1.0 / A2AToolboxTool / RemoteA2A pattern.

We verified:

  • The RemoteA2A connection is configured correctly.
  • A2AToolboxTool is created successfully.
  • The target Hosted Agent has an agent card and incoming A2A configuration.
  • The agent card is accessible.
  • The correct project identity has the required Foundry Agent Consumer role.
  • Authentication and connection resolution succeed.
  • The failure occurs when the actual A2A message is sent.

The platform returns: -32099 HOSTED_AGENT_NOT_SUPPORTED The requested A2A operation is not supported for hosted-agent target. Use a prompt agent as the A2A target.

We also tested direct A2A previously and received the same result. As another workaround, we tried calling the specialist's Responses endpoint directly. The request reaches the Hosted Agent, but its hosting framework requires a trusted Foundry platform user identity in the request context. Forwarding the user ID from the calling Hosted Agent does not populate the trusted context, resulting in an HTTP 500.

Question

Is Hosted → Hosted agent-to-agent delegation currently supported in Azure AI Foundry?

If not, what is the officially supported architecture for keeping multiple custom-code Hosted Agents as independently deployed agents while allowing one Hosted Agent to delegate work to another?

Is there another Foundry-managed mechanism, such as Workflows or another agent/tool composition mechanism, that supports:

Hosted Agent A -> independently deployed Hosted Agent B

without converting Agent B to a Prompt Agent or moving its code inside Agent A?

We want to confirm whether this is a platform limitation or whether there is a supported configuration/transport we are missing.

Foundry Agent Service
Foundry Agent Service

A fully managed platform in Microsoft Foundry for hosting, scaling, and securing AI agents built with any supported framework or model

0 comments

1 answer

Sort by: Oldest
Most helpful Newest Oldest
  1. Allan Solomon Mejia 10,225 Reputation points
    2026-09-25T14:23:32.8733333+00:00

    Hello @Manivannan M

    Based on Microsoft's current Foundry documentation, hosted agents can expose A2A endpoints. Hosted agents support A2A v1.0 (GA) and v0.3 (preview), provided the A2A protocol is declared in the hosted agent version configuration. (Microsoft Learn)

    The outbound A2A pattern for Foundry Agent Service is for an agent to connect to a remote A2A endpoint using the A2A tool. Therefore, first verify that each target Hosted agent was deployed with the A2A protocol explicitly enabled, rather than only Responses/Invocations, and that you are targeting its A2A endpoint:

    {project_endpoint}/agents/{name}/endpoint/protocols/a2a

    Hosted-agent endpoints are active depending on the protocols declared in the agent version definition.

    One important distinction: Microsoft's separate documentation for enabling incoming A2A on a Foundry agent currently lists a deployed prompt agent as the prerequisite for that specific configuration path.

    So there appears to be an important distinction between the Prompt-agent A2A configuration flow and the newer Hosted-agent protocol endpoint.

    I cannot find verified Microsoft documentation explaining error -32099 HOSTED_AGENT_NOT_SUPPORTED or stating that Hosted → Hosted A2A is intentionally unsupported. Given that the current Hosted-agent documentation explicitly exposes an A2A endpoint, do not treat that error alone as confirmation of an architectural limitation.

    If the target Hosted agents already have A2A enabled and direct invocation of their /protocols/a2a endpoints succeeds, but RemoteA2A/A2AToolboxTool still returns HOSTED_AGENT_NOT_SUPPORTED, I recommend escalating this to Microsoft with the source/target agent versions, protocol configuration, request/correlation ID, and the exact -32099 response.

    For an alternative architecture, you can use Hosted agent workflows, where multiple agents collaborate in a code-based workflow while retaining separate models, tools, and instructions. However, that is a different composition model and doesn't, by itself, establish that independently deployed Hosted → Hosted A2A should be replaced.

    References:

    Hosted agents in Foundry Agent Service

    Connect to an A2A agent endpoint

    Enable incoming A2A on a Foundry agent

    Create hosted agent workflows

    Help make this community better for everyone: If this answer helped or resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution. Thank you.

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.