- Runs it
- Jointly governed by Zed and JetBrains
- Projects
- 20
- Official
- 8
- Stars checked
- 2026-11-08
- Spec
- official
ACP Coding Agents, Adapters and SDKs: 20 open-source projects compared
Find coding agents that plug into Zed, JetBrains and Neovim through the Agent Client Protocol. Compare 20 open-source ACP agents, adapters and SDKs by stars, licence and recent activity.
Compare ACP projects
These 20 open-source projects implement or support ACP. 8 are maintained by the official project and the rest by the community. Sort by stars or recent activity, or filter by type. These projects are listed from public GitHub data and are not scanned by AgenticMart.
Stars and activity checked 2026-11-08. These projects are listed from public GitHub data and are not scanned by AgenticMart. Built one? List it free now, paid selling opens soon.
What is ACP?
The Agent Client Protocol (ACP) standardizes communication between code editors and coding agents, much as the Language Server Protocol did for language servers. Local agents run as editor sub-processes using JSON-RPC over stdio, and remote agents use HTTP or WebSocket. It is a different project from the Agent Communication Protocol, which shares the ACP abbreviation.
ACP vs MCP: what is the difference
ACP connects an editor to a coding agent, while MCP connects AI applications to data and tools. ACP reuses MCP's JSON formats where it can and adds its own types for coding features such as showing diffs. New to MCP? Read what an MCP server is.
What an ACP message looks like
The editor (the ACP client) sends the user's message to the coding agent inside an existing session as a JSON-RPC request. The prompt is a list of content blocks, here a text instruction plus the contents of a file the user attached.
{
"jsonrpc": "2.0",
"id": 2,
"method": "session/prompt",
"params": {
"sessionId": "sess_abc123def456",
"prompt": [
{
"type": "text",
"text": "Can you analyze this code for potential issues?"
},
{
"type": "resource",
"resource": {
"uri": "file:///home/user/project/main.py",
"mimeType": "text/x-python",
"text": "def process_data(items):\n for item in items:\n print(item)"
}
}
]
}
}From the official specification, checked 2026-11-12.
ACP questions
What is the Agent Client Protocol?
The Agent Client Protocol (ACP) is a standard way for code editors and AI coding agents to talk to each other. It follows the idea of the Language Server Protocol, so one agent can work in many editors without a custom integration for each. It is built on JSON-RPC 2.0.
ACP vs MCP: what is the difference?
ACP connects an editor to a coding agent, while MCP connects AI applications to data and tools. ACP reuses MCP's JSON formats where it can and adds its own types for coding features such as showing diffs.
Which editors support ACP?
As of November 2026, the ACP site lists Zed, JetBrains IDEs, Neovim, Emacs, Visual Studio Code and Sublime Text among its clients. Several of these, including Neovim, Emacs and VS Code, get support through plugins or extensions rather than built-in features.
How do I get started with ACP?
Pick an official SDK for Python, TypeScript, Rust, Kotlin or Java and follow the protocol overview on agentclientprotocol.com. The SDK helps you build either side, the agent or the client. You can test against an editor from the site's client list, and related SDKs and adapters are listed below.
How does the Agent Client Protocol work?
The editor starts or connects to an agent, opens a session and sends prompts, and the agent streams its progress back. Local agents run as editor subprocesses using JSON-RPC over stdio, and remote agents use HTTP or WebSocket. Agents can also ask the editor for file access, terminals and user permission for tool calls.
Who maintains the Agent Client Protocol?
Zed and JetBrains jointly govern the Agent Client Protocol as of November 2026. The project calls this an interim model and says it is working toward an independent foundation, without naming one or giving a date.
What are the limits of the Agent Client Protocol?
ACP's main limits as of November 2026 are unfinished remote agent support and an interim governance model. Its scope is the link between editors and coding agents. The project says it is moving toward an independent foundation, but it names no foundation or timeline.