Goose Adapter

Installation

pip install openharness-goose

Prerequisites

Goose must be running as a server:

# Install Goose
brew install block/tap/goose  # macOS

# Start the server (default port 3000)
goose server

Quick Start

from openharness_goose import GooseAdapter
from openharness.types import ExecuteRequest

# Connect to local Goose server
adapter = GooseAdapter(base_url="http://localhost:3000")

# Start a session with working directory
session_id = await adapter.start_session(
    working_directory="/path/to/project"
)

# Execute a prompt
result = await adapter.execute(
    ExecuteRequest(
        message="Help me understand this codebase",
        session_id=session_id
    )
)
print(result.output)

# Stop the session when done
await adapter.stop_session(session_id)

Capabilities

Supported

Not Supported

Session Management

Goose uses sessions to maintain conversation context and working directory:

# Start a new session
session_id = await adapter.start_session(
    working_directory="/my/project",
    recipe_name="developer"  # Optional pre-configured recipe
)

# List all sessions
sessions = await adapter.list_sessions()
for session in sessions:
    print(f"{session.id}: {session.name}")

# Resume an existing session
await adapter.resume_session(session_id)

# Export session (for backup/sharing)
export_data = await adapter.export_session(session_id)

# Import session
new_session_id = await adapter.import_session(json.dumps(export_data))

# Delete session
await adapter.delete_session(session_id)

MCP Extensions

Add powerful tools through Model Context Protocol servers:

from openharness_goose.types import GooseExtension

# Add a built-in extension
await adapter.add_extension(
    session_id,
    GooseExtension(
        name="developer",
        type="builtin",
    )
)

# Add an MCP server via stdio
await adapter.add_extension(
    session_id,
    GooseExtension(
        name="filesystem",
        type="stdio",
        cmd="npx",
        args=["-y", "@anthropic-ai/mcp-server-filesystem", "/tmp"],
    )
)

# Add an MCP server via SSE
await adapter.add_extension(
    session_id,
    GooseExtension(
        name="remote-tools",
        type="sse",
        uri="https://example.com/mcp/sse",
    )
)

# List extensions for a session
extensions = await adapter.get_session_extensions(session_id)

# Remove an extension
await adapter.remove_extension(session_id, "filesystem")

Extension Types

Streaming

from openharness.types import ExecuteRequest

async for event in adapter.execute_stream(
    ExecuteRequest(
        message="Explain this code step by step",
        session_id=session_id
    )
):
    if event.type == "text":
        print(event.content, end="")
    elif event.type == "tool_call_start":
        print(f"\n[Tool: {event.name}]")
    elif event.type == "tool_result":
        print(f"[Result: {event.output}]")
    elif event.type == "done":
        print("\n[Complete]")

Event Types

Multi-Model Support

Goose supports 25+ LLM providers:

# Switch to Anthropic
await adapter.update_provider(
    session_id,
    provider="anthropic",
    model="claude-3-5-sonnet-20241022",
)

# Switch to OpenAI
await adapter.update_provider(
    session_id,
    provider="openai",
    model="gpt-4o",
)

# Use local models via Ollama
await adapter.update_provider(
    session_id,
    provider="ollama",
    model="llama3.2",
)

Supported Providers

Anthropic, OpenAI, Google (Gemini), Azure OpenAI, AWS Bedrock, Ollama, Groq, Together AI, Mistral, Cohere, and many more.

Configuration

Adapter Options

adapter = GooseAdapter(
    base_url="http://localhost:3000",  # Goose server URL
    timeout=60.0,                        # Request timeout in seconds
)

Starting Goose Server

# Default settings
goose server

# Custom port
goose server --port 8080

# With working directory
goose server --working-dir /my/project

Provider Configuration

Configure LLM providers through Goose CLI:

goose configure