Section 7 · Multi-Agent Orchestration · 5% of exam
7.5 Agent-to-Agent (A2A) Communication
Exam objective
Explain how the A2A protocol supports communication and task delegation between agents and distinguish it from MCP.
Standardize communication between agents
Agent-to-Agent, or A2A, is an open protocol intended to let agents communicate and collaborate across different platforms or frameworks. It provides a shared way for one agent to discover another agent’s capabilities, send a request, and receive a result or task update. The goal is interoperability: a calling agent can delegate work to an appropriately described remote agent without needing to know its internal model, prompts, or implementation details. A2A addresses communication between agents, while leaving each agent responsible for how it reasons and performs its own work.
Use capability descriptions to route work
An A2A agent may publish an Agent Card, a machine-readable description of its identity, capabilities, skills, endpoint, and authentication requirements. A client can use that description to decide whether the agent is suitable, then communicate using the protocol’s message and task patterns. For example, an agent responsible for travel policy might delegate a narrow eligibility question to a remote benefits agent and use the returned answer in its own workflow. The capability description should be accurate and the request should contain only the information needed for the task.
Keep A2A distinct from MCP
MCP and A2A are complementary but answer different questions. MCP standardizes how an AI application accesses tools, data, and other server-provided capabilities. A2A standardizes how agents communicate and delegate tasks to other agents. A simple memory aid is “MCP connects an agent to tools; A2A connects agents to agents.” A remote agent may use its own tools behind the A2A interface, but the caller does not need to invoke those tools directly. In a Salesforce exam scenario, choose MCP when the need is an external function or data source and A2A when the need is agent collaboration or interoperability.
Treat interoperability as a governed boundary
A standard protocol does not mean every agent should be trusted automatically. Before delegation, evaluate the remote agent’s identity, advertised capability, authentication, data handling, response format, and failure behavior. Decide what information may cross the boundary and whether a human review or validation step is needed before acting on the response. Test success, refusal, timeout, malformed output, and unexpected content. A2A can help systems communicate consistently, but business owners still need to define authority and access. Verify which protocol features a specific Salesforce product or release supports instead of assuming that an open standard guarantees every integration path is available.
Remember the distinction: MCP is for tool and system connections; A2A is for communication and delegation between agents.
An open standard defines a common communication method. It does not establish identity, permission, correctness, or trust by itself.
Extra concept: A protocol can hide an agent’s internal implementation while still exposing its declared capabilities and the result of its work.
Official references:
- Salesforce Certified Agentforce Specialist Exam Guide https://help.salesforce.com/s/articleView?id=005298924&language=en_US&type=1
- Agent2Agent Protocol Specification https://a2a-protocol.org/latest/specification/
- Announcing the Agent2Agent Protocol https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/
Key takeaways
- A2A standardizes agent-to-agent communication and task delegation across implementations.
- Agent Cards can describe an agent’s capabilities, endpoint, and authentication requirements.
- MCP connects agents to tools and systems; A2A connects agents to other agents.
