Claude Code Adapter

Installation

pip install openharness-claude-code

Prerequisites

The Claude Agent SDK requires the Claude Code CLI to be installed:

# macOS/Linux/WSL
curl -fsSL https://claude.ai/install.sh | bash

# Or via Homebrew (macOS)
brew install --cask claude-code

# Or via WinGet (Windows)
winget install Anthropic.ClaudeCode

Authentication

Set your Anthropic API key:

# Required: Set your API key
export ANTHROPIC_API_KEY="your-api-key"

# Alternative: Use cloud providers
# Amazon Bedrock
export CLAUDE_CODE_USE_BEDROCK=1  # + AWS credentials

# Google Vertex AI
export CLAUDE_CODE_USE_VERTEX=1   # + GCP credentials

# Microsoft Azure
export CLAUDE_CODE_USE_FOUNDRY=1  # + Azure credentials

Get your API key from the Anthropic Console. For complete setup instructions, see the official Claude Agent SDK documentation.

Quick Start

from openharness_claude_code import ClaudeCodeAdapter, ClaudeCodeConfig
from openharness.types import ExecuteRequest

# Create adapter with configuration
config = ClaudeCodeConfig(
    cwd="/path/to/project",
    model="sonnet",
    permission_mode="acceptEdits",
)
adapter = ClaudeCodeAdapter(config)

# Execute a prompt with streaming
async for event in adapter.execute_stream(
    ExecuteRequest(message="Explain this codebase")
):
    if event.type == "text":
        print(event.content, end="")

# Or execute synchronously
result = await adapter.execute(
    ExecuteRequest(message="What does the main function do?")
)
print(result.output)

Capabilities

Supported

Not Supported

Built-in Tools

Claude Code provides 12 built-in tools for file operations, web access, and agent orchestration:

Listing Tools

tools = await adapter.list_tools()
for tool in tools:
    print(f"{tool.name}: {tool.description}")

Streaming

from openharness.types import ExecuteRequest

async for event in adapter.execute_stream(
    ExecuteRequest(message="Refactor this function")
):
    if event.type == "text":
        print(event.content, end="")
    elif event.type == "thinking":
        print(f"[Thinking: {event.thinking}]")
    elif event.type == "tool_call_start":
        print(f"\n[Tool: {event.name}]")
    elif event.type == "tool_result":
        print(f"[Result: {event.output[:100]}...]")
    elif event.type == "done":
        print(f"\n[Tokens: {event.usage.total_tokens}]")

Event Types

MCP Integration

Add custom tools via Model Context Protocol servers:

config = ClaudeCodeConfig(
    cwd="/my/project",
    mcp_servers={
        # stdio server (local subprocess)
        "filesystem": {
            "command": "npx",
            "args": ["-y", "@anthropic-ai/mcp-server-filesystem", "/tmp"],
        },
        # SSE server (remote)
        "database": {
            "uri": "https://my-mcp-server.com/sse",
        },
    },
)
adapter = ClaudeCodeAdapter(config)

Configuration

ClaudeCodeConfig Options

config = ClaudeCodeConfig(
    model="sonnet",               # Model: sonnet, opus, haiku
    cwd="/path/to/project",       # Working directory
    system_prompt="Be helpful",   # Custom system prompt
    allowed_tools=["Read", "Write"],  # Restrict to specific tools
    permission_mode="acceptEdits",   # Permission mode
    mcp_servers={...},            # MCP server configuration
    max_turns=10,                 # Maximum agent turns
    env={"DEBUG": "true"},        # Environment variables
)

Permission Modes

Available Models