Goose

Goose is an open-source AI agent that runs on your own machine as a desktop app, a command-line tool, and an embeddable API. It goes beyond suggesting code: it installs dependencies, executes commands, edits files, and tests its work locally. It extends through the Model Context Protocol , so the same agent can reach databases, APIs, browsers, and services through a growing set of MCP extensions. It is released under the Apache-2.0 license and is written mostly in Rust.

Goose was created by Block and is now governed by the Agentic AI Foundation (AAIF) at the Linux Foundation, a vendor-neutral, community-run home. The project moved to the aaif-goose organisation, and its documentation now lives at goose-docs.ai. It is a general-purpose agent rather than a coding-only tool, which makes it useful for research, data work, and automation as well as writing software.

Where Goose sits in the stack

Installation

Goose installs as a command-line tool through an official script. The desktop app is a separate download from the releases page.

curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash

After installing, configure a model provider before your first run.

goose configure    # choose "Configure Providers", pick a provider, enter an API key

Two ways to use it

The first pattern is an interactive session. You start a session, work with the agent, and resume it later where you left off.

goose session        # start a new session
goose session -r     # resume the previous session

The second pattern is adding an MCP extension so Goose can reach a new system. You can add one interactively, or load it directly when starting a session.

# load a built-in extension and an MCP server at launch
goose session --with-builtin "developer" \
              --with-extension "uvx mcp-server-fetch"

For repeatable work, Goose captures a task as a recipe: a portable YAML file you can share with your team and run in CI. A minimal recipe needs a title, a description, and either instructions or a prompt.

version: "1.0.0"
title: "Summarize open issues"
description: "Fetch and summarize this week's open GitHub issues"
prompt: "List open issues opened in the last 7 days and group them by label"
settings:
  goose_provider: "anthropic"

Typical workflow

How it compares

When not to use it

• You want a managed cloud service. Goose runs on your machine, so you own setup, secrets, and sandboxing. If you want a hosted agent with no local footprint, a cloud offering suits better.
• You only want one model in one terminal. Goose is a broad, general-purpose agent. For a narrow single-vendor coding CLI, Claude Code or a lighter tool is less overhead.
• You cannot grant local execution. Goose installs and runs commands on your machine. In environments where an agent cannot have that access, it is the wrong fit.
• You need per-edit approval in an editor. For a reviewed, diff-by-diff IDE workflow, Cline is designed around that gate.

Further reading

• Goose documentation : official install, providers, extensions, and recipes.
• Goose on GitHub : source and releases under the Apache-2.0 license.
• Agentic AI Foundation : the Linux Foundation home that now governs Goose.
• Open-source coding agents compared : where Goose sits among the alternatives.
• What is the Model Context Protocol? : the standard Goose uses for extensions.

Sources

• Goose documentation : definition, install, session and provider configuration, extensions, and recipes.
• Goose on GitHub : Apache-2.0 license, Rust, and the move from Block to the aaif-goose organisation.
• Linux Foundation announces the Agentic AI Foundation : the governance move for Goose.