Agent Client Protocol
lastmod 2025-09-03created 2025-09-031 edits
Agent Client Protocol (ACP)
About
- Agent Client Protocol (ACP) is a protocol for standardizing communication between code editors and AI coding agents
- Previously, each editor needed separate custom integration work to connect to a specific agent, which caused limited compatibility and developer lock-in
- Like the Language Server Protocol (LSP), ACP provides a standardized way to communicate, so that once implemented, any compatible editor and agent can connect freely
- The editor runs the agent as a subprocess and communicates via JSON-RPC over stdio; UI elements support diff display and Markdown-based output
- Currently Zed, neovim (CodeCompanion plugin), and others support ACP; on the agent side, Gemini CLI is compatible, and support is expected to expand
Introduction
- Agent Client Protocol (ACP) is an open protocol designed to standardize server-client interactions such as remote development, port forwarding, and command execution
- Existing problem: AI coding agents and editors are tightly coupled, but interoperability is not guaranteed by default
- Each editor must build a custom integration for every agent it wants to support
- Agents must implement a specific editor's API to reach users
- This causes integration overhead, limited compatibility, and developer lock-in
- An agent that implements ACP works in every compatible editor
- An editor that supports ACP can access the entire ecosystem of ACP-compatible agents
- Enables independent innovation and lets developers choose the tools best suited to their workflow
ACP overview
- How it works: users mainly work in a code editor and invoke an agent for specific tasks
- The agent runs as a subprocess of the editor
- Communicates using JSON-RPC over stdio
- Rich formatting is possible without HTML rendering
Current support
- Editors
- Zed: supports ACP, per the official documentation
- Neovim: ACP integration via the CodeCompanion plugin
- Support for more editors planned
- Gemini: supports ACP via Gemini CLI
- Support for more agents planned
Conclusion
- Modeled on the success of LSP, ACP dramatically improves interoperability between AI coding agents and editors
- Developers can freely choose the best combination of tools without being locked into a specific agent or editor
- Extending the protocol increases the ecosystem's scalability and can improve the efficiency and flexibility of developer workflows