Goose Engine

Block's open-source agentic AI tool. Goose is a fully autonomous CLI agent that extends its capabilities through MCA (Multi-Chain Agents) tool extensions — all surfaced through QuantumReef's unified session UI.

Last updated: 2025-01-15

Overview

Goose is an open-source autonomous AI agent created by Block (formerly Square). Unlike prompt-and-respond tools, Goose operates in fully autonomous loops — it plans a task, executes steps via tools, observes results, and iterates until the goal is achieved. QuantumReef integrates Goose via its --non-interactive mode, streaming its structured output into the session timeline and tools panel.

Goose's defining feature is its MCA (Multi-Chain Agents) extension system. Extensions add new tool capabilities — from browser automation to database access — without modifying Goose's core. QuantumReef's MCP Tools Panel displays active MCA extensions alongside their invocations in real time.

Open-source and provider-agnostic

Goose is fully open-source (Apache 2.0) and supports any OpenAI-compatible backend — Anthropic, OpenAI, Ollama, and more. No single-vendor lock-in.

Installation

Install Goose

bash

# Install via pip
pip install goose-ai

# Or via pipx (isolated environment — recommended)
pipx install goose-ai

# Verify installation
goose --version

Configure your LLM provider

bash

# Anthropic (recommended for complex tasks)
export ANTHROPIC_API_KEY=sk-ant-...

# OpenAI
export OPENAI_API_KEY=sk-...

# Local Ollama (no key required)
# Ensure Ollama is running: ollama serve
# Then set provider to "ollama" in goose.yaml

Enable in QuantumReef

Navigate to Settings → Engines → Goose and toggle on. QuantumReef detects the goose binary from your PATH. For pipx installations, the binary is typically at ~/.local/bin/goose.

Configuration

goose.yaml

Goose reads its configuration from ~/.config/goose/goose.yaml. A project-level .goose/goose.yaml overrides the global config for that workspace.

~/.config/goose/goose.yaml

# LLM provider and model
provider: anthropic
model: claude-opus-4-5

# Fallback model for lightweight steps
weak_model: claude-haiku-3-5

# Autonomous loop settings
max_iterations: 20
stop_on_error: false

# MCA extensions to load
extensions:
  - developer    # file R/W, shell execution, git
  - browser      # headless browser via Playwright
  - memory       # persistent cross-session memory

# Tool permissions
allow_shell: true
allow_write: true
allow_network: true

QuantumReef Engine Settings

Setting	Default	Description
Provider	anthropic	LLM backend: anthropic, openai, ollama, etc.
Model	claude-opus-4-5	Primary model for agentic reasoning
Max Iterations	20	Maximum autonomous loop iterations before halting
Extensions	developer	Comma-separated MCA extensions to load
Allow Shell	true	Permit Goose to run shell commands
Stop on Error	false	Halt the loop on the first tool error

MCA Tool Extensions

MCA extensions are the primary way to extend Goose's capabilities. Each extension adds a set of tools the agent can invoke during its autonomous loop. QuantumReef shows all active extensions and their tool calls in the MCP Tools Panel.

Common MCA extensions for Goose
Extension	Install	Provides
developer	Built-in	File R/W, shell commands, git operations, code search
browser	pipx inject goose-ai goose-browser	Playwright-powered headless browser, scraping, screenshots
memory	pipx inject goose-ai goose-memory	Persistent key-value memory across sessions
github	pipx inject goose-ai goose-github	PR management, issue creation, code review comments
jira	pipx inject goose-ai goose-jira	Issue tracking, sprint management, comment threads

bash

# Install the browser extension
pipx inject goose-ai goose-browser

# List installed extensions
goose extensions list

# Run Goose with specific extensions enabled
goose run --extensions developer,browser "Scrape the docs site and generate a summary"

When to Choose Goose

Use Case	Why Goose Excels
Fully autonomous tasks	Set-and-forget loop runs without human intervention
Multi-step workflows	Iterates through plan → execute → observe cycles automatically
Browser automation	Browser MCA extension drives real web pages via Playwright
Open-source teams	Apache 2.0 license, full source available, self-hostable
Multi-provider teams	Swap between Anthropic, OpenAI, and Ollama via one config line

Best for long-running tasks

Goose shines when you want to hand off a complex, multi-step task and walk away. Set max_iterations high and let it loop until done. QuantumReef's session timeline shows each iteration step as it completes.