What Is goose (goose-docs.ai)?

goose is an open-source, general-purpose AI agent that runs on your own machine. It is not limited to coding: the same agent can handle research, writing, automation, and data analysis. You can use it through a native desktop app (macOS, Linux, Windows), a command-line interface, or an API you embed in your own software. It is built in Rust for performance and portability, works with 15+ LLM providers, and connects to 70+ extensions through the Model Context Protocol (MCP). If you want a local, extensible agent you can point at your own tools and models, goose is worth evaluating; if you need a fully managed cloud service with no local setup, it is a different category of tool.

What goose actually does

goose is a general-purpose agent, not a single-purpose coding assistant. According to the project's own description, it is meant for "research, writing, automation, data analysis, or anything you need to get done." The practical implication is that the same agent runtime can be pointed at different tasks by changing the extensions, skills, and instructions you give it.

Three delivery surfaces are offered:

Because all three share the same underlying agent, a workflow you prototype in the desktop app can generally be moved to the CLI or driven through the API.

How it connects to tools and models

Extensions via MCP

goose connects to external systems through the Model Context Protocol, an open standard for linking AI agents to tools and data sources. The project describes itself as one of the earliest adopters of MCP and claims one of the deepest integrations in the ecosystem, with 70+ documented extensions covering databases, APIs, browsers, GitHub, Google Drive, and more. You can also add skills or build your own extension.

The practical benefit: instead of the agent being hard-wired to a fixed set of tools, you choose which capabilities to attach per task.

Any LLM, including existing subscriptions

goose works with 15+ providers, including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, and Bedrock. You can supply API keys, or use existing Claude, ChatGPT, or Gemini subscriptions through ACP (Agent Client Protocol). That means you are not locked into a single model vendor, and you can mix local models (via Ollama) with hosted ones.

ACP in both directions

goose acts as an ACP server, so you can connect to it from Zed, JetBrains, or VS Code. It can also use ACP agents such as Claude Code and Codex as providers. In other words, goose can be the agent your editor talks to, or it can sit on top of another agent.

Recipes, subagents, and MCP Apps

• Recipes — Capture a workflow as a portable YAML config that includes instructions, extensions, parameters, and subrecipes. Recipes can be shared with a team or run in CI, which is what makes repeatable automation practical rather than ad hoc.
• Subagents — Spawn independent subagents to handle tasks in parallel, such as code review, research, or file processing, keeping the main conversation clean.
• MCP Apps — Extensions can render interactive UIs directly inside goose Desktop, including buttons, forms, and visualizations.

Security features

goose lists several safety mechanisms: prompt injection detection, tool permission controls, sandbox mode, and an adversary reviewer that watches for unsafe actions. These matter most when you give the agent access to real systems (files, databases, APIs), since an agent with broad tool access inherits the risk of those tools.

Governance and community

goose is part of the Agentic AI Foundation at the Linux Foundation, which the project says keeps it vendor-neutral, community-governed, and open for the long term. Community channels include Discord (questions and help), GitHub (stars, forks, issues, contributions), and a browsable catalog of community-built MCP extensions.

When goose is a good fit — and when it isn't

Consider goose if you:

• Want an agent that runs locally on macOS, Linux, or Windows
• Need to connect an agent to your own databases, APIs, or internal tools via MCP
• Want to switch between LLM providers, or use a local model through Ollama
• Need repeatable, shareable workflows (Recipes) that can run in CI
• Care about open governance and an open-source codebase

Look elsewhere if you:

• Want a fully hosted service with no local installation or configuration
• Need a narrowly scoped coding-only assistant and nothing else
• Cannot run a local agent in your environment for policy or hardware reasons

Getting started

The project's own entry points are the "Get started" section and a "See goose in action" demo on goose-docs.ai, plus the extension browser and community links. The fastest path is usually: install the desktop app for your OS, connect a model provider (API key or existing subscription), then add the extensions you need for your first task. From there, capture anything you repeat as a Recipe so it can be shared or run in CI.

Note that the source material does not state pricing or login requirements, so check the site directly before assuming goose is free of cost or sign-in for your intended use.