Goose
Open source, extensible AI agent that goes beyond code suggestions
Install with Homebrew
brew install --cask block-gooseQuick Take: Goose
Goose remains one of the most flexible open agent runtimes on Mac in August 2026: free, MCP-native, multi-model, and now foundation-governed under AAIF. Coding specialists may prefer Claude Code, Cursor, or Muse Code for pure edit loops, but if you want an agent that can grow with your tools—not just your editor—Goose is still the open default. Pair it with strong permission hygiene and, for teams, OpenTelemetry.
Best For
- •Platform engineers automating multi-tool workflows
- •Teams standardizing on MCP
- •Developers who want local-first agents with model choice
What is Goose?
Goose is an open-source, extensible AI agent stewarded by the Agentic AI Foundation (AAIF) at the Linux Foundation. Block (formerly Square) created Goose and donated it to AAIF alongside Anthropic's Model Context Protocol (MCP) and OpenAI's AGENTS.md, so the project is no longer a single-vendor side project—it is foundation-hosted infrastructure for local agents. The primary GitHub home is now aaif-goose/goose; the desktop app, CLI, and API still ship under the Apache 2.0 license. Goose is not a code editor or a chat box bolted onto an IDE. It is a general-purpose agent that runs on your machine: install tools, execute shell commands, edit files, browse, manage git, and orchestrate multi-step workflows. Out of the box it reads and writes files, runs commands, works with GitHub, and talks to external services through MCP extensions. With 70+ documented extensions and support for 15+ model providers—including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock, plus 2026 additions such as Alibaba Qwen (DashScope), Perplexity, Databricks AI Gateway, Vercel AI Gateway, xAI SuperGrok subscriptions, Scaleway, and NEAR AI Cloud—you choose the model and the tool surface. As of late July 2026 the project was releasing on the v1.45 line (v1.45.0 landed 2026-07-29), with frequent point releases through the summer. June 2026's 1.36/1.37 pair doubled down on open provider integrations so teams are not locked to a single frontier model. You can also attach existing Claude, ChatGPT, or Gemini subscriptions through the Agent Client Protocol (ACP) instead of minting separate API keys for every workflow. On macOS, Goose ships a native desktop app plus a CLI installable via the official download script or Homebrew casks (block-goose / block-goose-cli remain the familiar brew tokens for many machines). It targets Apple Silicon and modern macOS, including Tahoe 26.x workflows. Competitive pressure in August 2026 is intense: Claude Code, OpenAI's Codex surface inside the ChatGPT desktop app, Cursor, and Meta's newly launched Muse Code (August 2026) all compete for agent mindshare. Goose's differentiator remains breadth—MCP extensibility, local-first execution, foundation governance, and model agnosticism—rather than a single closed coding loop. When evaluating whether to standardize on this app for a Mac-based workflow in late 2026, weigh three practical factors: how often the vendor ships security updates for current macOS releases including Apple Silicon and Tahoe 26.x, whether pricing is predictable (one-time, free open source, or subscription), and whether the tool still matches how your team actually works day to day. Documentation quality, export paths, and community responsiveness often matter more than feature checklists. Prefer official download channels or Homebrew casks/formulae, verify notarization where applicable, and keep a rollback plan when a major version changes storage formats or default settings. For AI-enabled products, re-check model names and rate limits quarterly—they change faster than UI chrome. For open-source tools, pin versions in Brewfiles or internal catalogs so a surprise upgrade does not break CI. Finally, capture a short internal note on why you chose the app: future you will thank present you when a flashy rival appears and you need a reasoned comparison instead of vibes.
Key Features
MCP extension system
Goose's core power is deep Model Context Protocol support. Connect databases, browsers, GitHub, cloud APIs, and internal tools through 70+ documented extensions. Extensions can render interactive UI inside Goose Desktop—buttons, forms, visualizations—so agents become task-specific tools rather than pure chat. Custom extensions plug the same protocol into proprietary systems.
Autonomous multi-step execution
Describe outcomes like reviewing open PRs ready to merge, and Goose navigates GitHub, reads diffs, checks CI, and returns structured summaries without hand-holding each click. It executes shell commands, edits files, and chains tools while showing what it is doing.
Multi-provider model support
Fifteen-plus providers cover Anthropic Claude (including current Sonnet 5 / Opus 5 generations when your keys allow), OpenAI GPT-5.6 family models, Google Gemini, Ollama local models, OpenRouter, Azure, Bedrock, and 2026 gateway additions (Qwen/DashScope, Perplexity, Databricks, Vercel AI Gateway, xAI SuperGrok, Scaleway, NEAR AI Cloud). ACP lets you reuse existing Claude, ChatGPT, or Gemini subscriptions.
Desktop, CLI, and API
Rust-built desktop apps for macOS, Linux, and Windows sit beside a full CLI and an embeddable API. Configurations and extensions stay shared so you can move between visual and terminal workflows without rewiring tools.
Recipes and subagents
Capture workflows as portable YAML Recipes with parameters and sub-recipes for CI or team reuse. Spawn independent subagents for parallel review, research, or file processing while keeping the main conversation readable.
Security, permissions, and observability
Permission controls, sandbox modes, prompt-injection awareness, and step-level transparency show which commands and files the agent touches. AAIF-era guidance also documents optional OpenTelemetry instrumentation so teams can audit agent actions in production.
Foundation governance (AAIF)
Since April 2026 Goose lives under the Agentic AI Foundation with Linux Foundation governance, alongside MCP and AGENTS.md. That matters for enterprises evaluating long-term stewardship beyond Block's original open-source release.
Who Should Use Goose?
1Platform Engineer
A platform engineer asks Goose to check microservice health, recent deployments, and elevated error rates. Goose queries Kubernetes, monitoring dashboards, and Slack incident channels via MCP extensions, then returns a status brief in about a minute.
2Full-Stack Developer
A developer describes a feature: open a branch, scaffold an API endpoint and React component, and open a draft PR. Goose coordinates git, file edits, and GitHub in one conversation while the developer reviews diffs.
3Open Source Maintainer
Each morning Goose triages new issues—labeling bugs vs questions, flagging duplicates, and drafting first responses. The maintainer approves or edits, turning a long triage block into a short review.
Install Goose on Mac
Goose is free and open source. Install the desktop app from official docs, use the CLI bootstrap script, or install via Homebrew casks if available on your machine. You need an LLM provider API key, an ACP-linked subscription, or a local Ollama model.
Install Homebrew (optional)
If you prefer brew: /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
Install Goose Desktop
Either download from goose-docs.ai installation docs, or run brew install --cask block-goose when the cask is available on your Homebrew setup.
Install Goose CLI (optional)
CLI: curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash or brew install block-goose-cli.
Configure a model provider
Launch Goose or run goose configure. Add Anthropic, OpenAI, Google, Ollama, OpenRouter, or another supported provider. Prefer local Ollama for sensitive codebases.
Configuration Tips
Start with built-in tools only
Enable filesystem, shell, git, and GitHub first. Add MCP extensions only after you trust the agent's permission prompts.
Prefer local models for secrets
Point Goose at Ollama when code cannot leave the machine. Use cloud frontier models only for non-sensitive tasks.
Turn on observability in team setups
For shared agent use, enable OpenTelemetry so destructive actions are auditable—AAIF community posts stress this is optional but important.
Alternatives to Goose
Goose competes with coding agents and general tool-using assistants. August 2026 rivals include foundation and commercial options.
Claude Code
Claude Code is Anthropic's terminal coding agent optimized for deep repository work with Claude models. Goose is broader—coding plus arbitrary MCP tools and multi-provider choice. Pick Claude Code for pure coding depth; Goose for extensible workflow orchestration.
Muse Code
Meta's Muse Code launched in August 2026 as a terminal coding agent on Muse Spark. It is a new low-cost rival for agent coding. Goose remains more MCP-centric and foundation-governed with desktop + CLI surfaces.
ChatGPT / Codex
OpenAI folded agentic coding into the ChatGPT desktop app (Codex merged mid-2026). ChatGPT is excellent for conversational coding inside OpenAI's ecosystem; Goose stays local-first and provider-agnostic.
Cursor
Cursor is an AI IDE with Composer models and deep editor integration. Goose is not an IDE—it is an agent runtime you extend. Use Cursor for day-long editing sessions; Goose for multi-tool automation outside a single editor.
Pricing
Goose is free under Apache 2.0. There is no Goose subscription. Costs are whatever your chosen LLM provider charges (or zero with local Ollama). Desktop, CLI, and API are included without paid seats.
Pros
- ✓Free, Apache 2.0, foundation-hosted at AAIF
- ✓MCP-first extensibility across 70+ extensions
- ✓Multi-provider models including local Ollama
- ✓Desktop + CLI + API on macOS Apple Silicon
- ✓Recipes and subagents for repeatable workflows
- ✓Transparent step-level tool execution
Cons
- ✗You manage LLM API keys and spend yourself
- ✗Custom extensions require engineering effort
- ✗UX is less polished than commercial IDEs like Cursor
- ✗Competitive coding agents ship faster productized loops
Community & Support
Goose community hubs include the aaif-goose/goose GitHub repository, Discord, and @goose_oss on X. AAIF publishes governance and release notes; Block engineers remain active contributors. Expect extension PRs, provider integrations, and observability guides rather than commercial support SLAs. Release cadence through mid-2026 has been rapid (point releases into the 1.4x series), so pin versions in CI when you depend on specific provider behavior. Discord remains the fastest place for extension troubleshooting; GitHub Issues is better for reproducible bugs.
Frequently Asked Questions about Goose
About the Author
Expert Tips for Goose
X and interview coverage of Block's internal Goose usage (Lenny Rachitsky with Block CTO) emphasizes multi-hour weekly savings and non-engineering teams adopting agents—treat that as enterprise proof, not consumer polish.
AAIF posts in August 2026 push OpenTelemetry for goose: agents make hard-to-explain destructive choices, so turn observability on before wide rollout.
Practitioners still recommend starting with filesystem/shell/git extensions only; MCP sprawl without permissions is how agents go sideways.
Related Technologies & Concepts
Related Topics
Sources & References
Key Verified Facts
- Goose is hosted by the Agentic AI Foundation at the Linux Foundation after Block's donation.[cite-2, cite-4]
- Primary source repository is aaif-goose/goose with desktop, CLI, and API surfaces.[cite-1]
- Releases 1.36/1.37 added multiple new model providers including xAI SuperGrok and Qwen via DashScope.[cite-3]
- Goose remains Apache 2.0 open source with no product subscription fee.[cite-5, cite-1]
- Muse Code emerged in August 2026 as a new macOS/Linux coding-agent competitor.[cite-7]
- 1aaif-goose/goose GitHub repository
Accessed Aug 9, 2026
"Official repo: open-source agent, desktop/CLI/API, AAIF membership, multi-provider support."
- 2goose moves to AAIF
Accessed Aug 9, 2026
"April 2026 blog: Block donated goose to AAIF at the Linux Foundation alongside MCP and AGENTS.md."
- 3goose Doubles Down on Open in Latest Two Releases
Accessed Aug 9, 2026
"June 2026 AAIF blog on 1.36/1.37 adding nine provider integrations including xAI SuperGrok and Qwen."
- 4TechCrunch: AAIF formation
Accessed Aug 9, 2026
"AAIF launch coverage: Anthropic MCP, Block Goose, OpenAI AGENTS.md donated to Linux Foundation effort."
- 5Block introduces codename goose
Accessed Aug 9, 2026
"Original Block announcement of open-source goose under Apache 2.0."
- 6Goose documentation home
Accessed Aug 9, 2026
"Historical docs portal defining goose as a general-purpose extensible agent."
- 7Meta launches Muse Code
Accessed Aug 9, 2026
"August 2026 competitor: Muse Code terminal agent for macOS/Linux."