Stacks and Flows

Protocol Boundary Guide · Source-verified

A2A vs MCP: different protocol boundaries

A2A describes communication between agents; MCP describes a model-facing context and tool boundary. They can coexist, but neither protocol removes application-level authorization, reliability, or compatibility work.

Protocol boundary

Two protocols, two boundaries

Agent A
A2A · delegate and coordinate
Agent B
MCP · request tools and context
MCP server
read / act
Tool / data source
MCP connects an agent to tools and context. A2A coordinates work between agents. They can sit in the same stack.

What the sources establish

  • The A2A project announced version 1.0 in March 2026 and AAIF Growth Stage status in August 2026.
  • A2A defines agent-to-agent concepts and operations in its versioned specification.
  • MCP maintains a separate specification and roadmap for connecting model applications to context, tools, and related capabilities.

Draw the boundary before choosing

A2A addresses communication across agent boundaries. MCP addresses how model applications connect to tools and context. A system can need either, both, or neither.

What neither protocol supplies

Application authorization, durable execution, business acceptance rules, and incident ownership remain outside the protocol label.

Decision detail

  1. Choose A2A when interoperating between independently operated agents is the actual requirement.
  2. Choose MCP when the boundary is between a model application and tools or context providers.
  3. Require version and implementation compatibility evidence before treating either protocol as portable infrastructure.
  4. Do not add either protocol when a direct application interface is the simpler boundary.

Evidence

These claims are source-verified. We do not label this page hands-on or benchmarked because no reproducible test artifact is attached.

Continue the decision