Integrating a Partner Agent with A2A in Microsoft Foundry
Short answer
B Adopt Agent-to-Agent (A2A) to integrate the partner risk-review agent with authenticated, discoverable, task-tracked calls that avoid coupling to its implementation.
Question
An orchestration agent
An external risk-review agent operated by a partner organization
The partner agent runs in a different framework, exposes its own endpoint, and manages long-running review tasks.
You need to identify an integration approach for the partner agent. The solution must meet the following requirements:
Use authenticated and auditable access for structured tools and data.
Support capability discovery and task tracking for the partner agent.
Prevent coupling the solution to the partner agent implementation.
What should you identify?
Key Takeaways
The test is whether you can separate agent-to-agent interoperability (A2A) from tool/data connectivity (MCP); the traps are MCP, which lacks agent discovery and task state, and in-process composition, which contradicts the partner's separate endpoint.
This AI-500 question tests how to integrate a partner-hosted risk-review agent running in a different framework. Agent-to-Agent (A2A) is the correct choice because it provides authenticated access, Agent Card capability discovery, and long-running task tracking without coupling to the partner's implementation.
Common mistake: Option D (Model Context Protocol) is the most plausible wrong answer because it offers authenticated access to structured tools and data, but MCP cannot discover a peer agent or track long-running tasks.
Expert Analysis
Reading the qualifiers The question asks you to "identify an integration approach" and then binds that choice with three must-meet requirements plus one architectural constraint. The decisive qualifiers are "capability discovery and task tracking for the partner agent" and "prevent coupling the solution to the partner agent implementation." Those two phrases point at a protocol-level, cross-organization contract rather than a code-level composition.
Why the partner's characteristics matter The partner agent runs "in a different framework," "exposes its own endpoint," and "manages long-running review tasks." An in-process option C is eliminated immediately: nothing that runs in-process can also be a separately hosted, independently versioned endpoint in another framework. Option A, platform-native subagent orchestration, is eliminated because native orchestration presumes the agents live inside the same Foundry platform boundary; a partner-operated service outside that boundary is exactly what native subagent wiring does not cover, and it would couple you to how the partner implements the agent.
MCP vs A2A Option D, MCP, is the trap. MCP does provide authenticated access to structured tools and data, which matches requirement one, but MCP has no notion of a peer agent, no capability discovery for an agent persona, and no task lifecycle for work that runs over minutes or hours. It answers "what can I call?", not "who is the peer and is my review task still running?" Requirement one is satisfied by A2A as well: A2A defines authentication/authorization between agent hosts and an auditable message exchange.
The correct choice Option B, Agent-to-Agent (A2A), satisfies all three at once: an Agent Card advertises capabilities (discovery), the task resource tracks states such as submitted/working/input-required/completed for long-running reviews (task tracking), and interaction happens over a published protocol rather than the partner's implementation details (decoupling), with authentication between the two agent hosts.
Selecting exactly one Only one option is asked for ("an integration approach"), and only B meets every listed requirement simultaneously.
Official Reference
https://a2a-protocol.org/latest/
This link opens the official vendor documentation in a new tab.
Exam Strategy
When a question pairs a technology with requirements, build a quick checklist and eliminate options that satisfy only part of it. Long-running tasks plus capability discovery plus cross-organization hosting is the signature of A2A; if the stem instead emphasized calling functions or reading files, MCP would dominate. Always check whether the question forbids in-process or platform-native coupling before selecting a composition option.
Frequently Asked Questions
When should I choose A2A instead of MCP in an AI-500 question?
Why is in-process agent-as-tool wrong for a partner-operated agent?
Does A2A provide authenticated and auditable access?
Community Votes
100% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.
Discussion 1 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.
From public community threads 1
-
ExamProAzure 👍 1 Answered BAnswer: B. Agent-to-Agent (A2A) A2A is the open protocol for one agent to communicate with another agent that is hosted elsewhere, which is exactly this scenario.
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.