Agent interoperability concept

What is Agent Client Protocol?

Reference

What is Agent Client Protocol? The Agent Client Protocol (ACP) is a protocol that lets code editors and coding agents communicate through a shared interface. In the ACP v2 documentation checked 2026-10-09, a client is usually an editor or another user interface, and an agent is a program that uses generative AI to modify code. ACP specifies how they exchange requests and updates; it is not the editor or agent itself.

Reviewed 2026-10-09

A common interface for editors and coding agents

The Agent Client Protocol is a communication protocol for code editors and coding agents. An editor is a program for viewing and editing source code; a coding agent is a program that uses generative AI to modify code. ACP gives those programs a shared way to communicate, so an editor can work with an agent that implements the protocol. In the ACP v2 documentation checked 2026-10-09, the client is typically an editor, but it can also be another user interface for interacting with an agent. The protocol describes the connection between the programs, not the programs themselves.

Requests and updates move between client and agent

The ACP v2 documentation checked 2026-10-09 describes a message system based on JSON-RPC 2.0. A method is a request that expects a result or error; a notification sends information without expecting a reply. A typical connection begins with initialization, where client and agent negotiate a protocol version and capabilities. The client can then create or resume a session and send a prompt. The agent can respond and send session updates about messages, state, and progress; it may also request permission for an operation. These are documented v2 interactions, and the exact supported capabilities depend on the client and agent.

It covers the editor-to-agent relationship

ACP fits the editor-agent interface part of agent interoperability: it describes how a coding environment communicates with an agent. A2A, in documentation checked 2026-10-09, describes a different relationship: communication and collaboration between AI agents. The two protocols therefore address different connections in an agent system. ACP is for the user-facing client and coding agent; A2A is for agents communicating with other agents. This distinction helps explain why ACP does not itself provide an agent’s coding abilities: it specifies how the client and agent communicate, while the connected programs provide their own functions.

ACP does not guarantee a particular feature set

ACP is not a coding agent, an editor, or a coding model; it is a protocol those programs can implement. In the ACP v2 documentation checked 2026-10-09, client and agent negotiate a protocol version, and capabilities describe the features each supports. The documentation says capabilities omitted during initialization are treated as unsupported. So, seeing that a program supports ACP does not by itself tell you which features it offers or what a particular client-agent pairing can do. Check the implementations’ supported versions and capabilities to understand that pairing.

Questions people ask

What does ACP stand for?

ACP stands for Agent Client Protocol. Its project documentation describes it as a protocol that standardizes communication between code editors and coding agents.

How do ACP coding agents communicate with an editor?

In the ACP v2 documentation checked 2026-10-09, the client initializes the connection, creates or resumes a session, and sends a prompt. The agent can respond and send session updates.

Is Agent Client Protocol the same as A2A?

No. ACP describes communication between a client, often a code editor, and a coding agent. A2A documentation checked 2026-10-09 describes communication and collaboration between AI agents.

Learn more

What changed

Sources

Part of Agent interoperability: Match the interface to the user

Questions and comments