Proof of concept integrating DeepAgents with AWS Bedrock AgentCore using the http via websockets transport for Agent Client Protocol (ACP).

This project demonstrates how to deploy a DeepAgents-powered agent as an AWS Bedrock AgentCore service using the Agent Client Protocol (ACP) for client-agent communication.

• Agent Framework: DeepAgents with LangGraph-based agent orchestration
• Protocol: Agent Client Protocol (ACP) for standardized client-agent communication (using a fork with HTTP transport)
• Transport: Custom HTTP transport via WebSockets that enables remote ACP agents to be connected with ACP clients.
• Integration: deepagents-acp library bridging DeepAgents with ACP protocol

This project uses a fork of the agent-client-protocol that implements an HTTP transport layer via WebSockets. This transport implementation enables remote ACP agents to be connected seamlessly with ACP clients, allowing the agent to run on AWS infrastructure while clients connect over standard WebSocket connections.

The agent uses the deepagents_acp project to expose a DeepAgents agent through the ACP protocol, allowing clients to interact with the agent using standardized ACP messages over WebSocket connections.

• Python 3.11+
• uv package manager
• AWS credentials configured (for Bedrock access)
• AWS Bedrock AgentCore CLI (for deployment)

Install dependencies using uv:

uv sync

Start the AgentCore development server:

agentcore dev

This will:

• Start the agent on http://localhost:8080
• Expose WebSocket endpoint at ws://localhost:8080/ws
• Enable hot-reloading for development

In a separate terminal, run the test client:

uv run test/test_acp_client.py

The test client will:

• Connect to the agent via WebSocket
• Initialize the ACP protocol
• Create a session
• Open an interactive prompt for sending messages to the agent

Environment variables can be configured in the agent:

• ask_before_edits: Agent requests permission before making file changes
• auto: Agent automatically executes file operations without asking

.
├── src/
│   └── acp_agent_main.py     # Main agent implementation
├── test/
│   └── test_acp_client.py    # Test client for connecting to agent
├── .bedrock_agentcore.yaml   # AgentCore configuration
├── pyproject.toml            # Project dependencies
└── README.md

• agent-client-protocol: Protocol and SDK for agent-client communication
• deepagents: LangGraph-based agent framework
• deepagents-acp: Integration layer between DeepAgents and ACP
• bedrock-agentcore: AWS Bedrock AgentCore SDK
• langchain-aws: AWS Bedrock integration for LangChain

1. AgentCore Runtime provides managed WebSocket infrastructure at /ws:8080
2. ACPDeepAgentBedrock implements the ACP protocol with a LangGraph-based DeepAgents agent
3. WebSocketStreamAdapter bridges Starlette WebSocket with asyncio streams
4. AgentSideConnection handles ACP protocol (JSON-RPC routing, method calls, responses)
5. Client connects via WebSocket and communicates using ACP messages

See src/acp_agent_main.py:3-21 for detailed architecture documentation.

Deploy to AWS using AgentCore CLI:

agentcore deploy

This will:

• Package the agent code
• Deploy to AWS Bedrock AgentCore
• Configure WebSocket endpoints
• Set up IAM roles and permissions

• Agent Client Protocol - Protocol specification and Python SDK
• DeepAgents - LangGraph-based agent framework
• DeepAgents ACP - ACP integration for DeepAgents