Which Agent Protocol Do You Need?

## What this tool does The agent protocol landscape has several overlapping standards with similar-sounding names, and teams often pick the wrong one. This selector asks a few short questions about what you are connecting and recommends the protocol that fits, along with what it is for and who created it. ## The protocols it weighs - **MCP (Model Context Protocol):** connects one agent to tools, data, and prompts. Created by Anthropic, now under the Linux Foundation. Use it when your agent needs reliable access to external capabilities. - **A2A (Agent2Agent):** lets independent agents discover and collaborate with each other. Created by Google Cloud, donated to the Linux Foundation. Use it when separate agents need to coordinate. - **Function calling (tool use):** the model-level mechanism where one model invokes one tool you defined. Use it when a single application calls its own tools and you do not need cross-vendor interoperability. - **ACP (Agent Client Protocol):** the Zed standard for an editor or CLI to drive one coding agent. Use it for editor-to-agent integration. ## MCP and A2A are complementary A common misconception is that MCP and A2A compete. They operate at different layers: MCP is the agent-to-tools layer, A2A is the agent-to-agent layer. Many real systems use both, and the tool says so when both apply. ## How to use it Answer the questions and read the result card. It names the recommended protocol, explains what it is for, and notes when to combine protocols rather than choose one. ## Related For the full picture, read our guide to AI agent protocols, which compares MCP, A2A, and the two different ACPs side by side.