deepagents-acp

What it is and what it does

deepagents-acp is a connector that exposes Deep Agents (LangChain's AI agent framework) as ACP servers, allowing them to run inside text editors that support the Agent Client Protocol. It wraps a Deep Agent instance and handles the ACP protocol layer, letting you define custom agents with tools and system prompts, then connect them to editors like Zed over stdio. The package includes examples for building a basic agent, persisting sessions using LangGraph checkpointers, and switching between different LLM models mid-conversation without losing history.

The main use case is embedding AI coding assistants and general-purpose agents directly into your editor workflow. You define an agent with whatever tools you need (filesystem access, shell commands, or custom functions), wrap it with AgentServerACP, and point your editor's ACP configuration at it. The package handles session management, conversation replay on reconnect, and model switching through ACP's session config options. It depends on agent-client-protocol for the wire protocol, deepagents for the agent runtime, and python-dotenv for environment variable loading (typically for API keys).

Use it for:

• Embed a custom coding agent in Zed with filesystem and shell tools, using Anthropic's Claude models
• Run a general-purpose Deep Agent with custom tools in any ACP-compatible editor without building your own protocol layer
• Switch between different LLM providers or models (Claude, GPT-4) mid-session while preserving conversation history
• Persist agent sessions across editor restarts using a durable LangGraph checkpointer for long-running workflows
• Launch a prebuilt coding agent (dcode) as an ACP server with a single command instead of wiring custom agent code

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Bridges Deep Agents with the Agent Client Protocol (ACP) to run AI agents in ACP-compatible text editors like Zed, exposing agent capabilities over stdio.

Yes, if you want to run Deep Agents in an ACP-compatible editor like Zed. The package is actively maintained, has low install friction, uses a permissive MIT license, and requires only standard dependencies. It's still in alpha (Development Status 3), so expect potential API changes, but it's suitable for experimentation and early adoption. Install it if you're already using Deep Agents and want editor integration; skip it if you don't need ACP connectivity or prefer a different agent framework.

Install

deepagents-acp on PyPI

pip install deepagents-acp

uv add deepagents-acp

poetry add deepagents-acp

Installing deepagents-acp

Before you install

Active maintenance with a recent release (2 days old) and low install friction. Requires Python 3.11 or later and three runtime dependencies (agent-client-protocol, deepagents, python-dotenv), all of which are standard ecosystem packages.

License in practice

MIT license permits free use, modification, and distribution with minimal restrictions, making it suitable for both open-source and commercial projects.

Quickstart

pip install deepagents-acp

import asyncio
from deepagents_acp.server import AgentServerACP
from deepagents import create_deep_agent
from acp import run_agent

async def main():
    agent = create_deep_agent(system_prompt="You are helpful")
    server = AgentServerACP(agent)
    await run_agent(server)

asyncio.run(main())

Requires Python 3.11 or later; agent must be created with a LangGraph checkpointer if session persistence is needed; ACP-compatible editor (e.g. Zed) must be configured to point to the server command.

Verify before relying

• Whether the package works with all ACP-compatible editors or only tested ones like Zed
• Performance characteristics when running agents with large conversation histories or complex tool sets
• Whether session persistence works reliably across different checkpointer implementations

Package facts

Evidence: deepagents_acp-0.0.10-py3-none-any.whl

Keywords: acp, agent, agent-client-protocol, ai-agents, deepagents

Tags

More Python Modules packages

Further reading