Goose — Open-Source AI Dev Agent by Block

What it is

Goose is an open-source AI agent built by Block (formerly Square) that goes beyond code suggestions. It installs packages, executes shell commands, manages files, and orchestrates entire developer workflows through a plugin system. The agent supports the Model Context Protocol (MCP) for connecting to external tools and data sources.

Goose targets developers who want an AI assistant that takes action rather than just generating text. If you need an agent that can set up projects, run tests, manage infrastructure, and interact with APIs, Goose provides the extensibility framework to do it.

How it saves time or tokens

Instead of copying AI-generated commands and pasting them into your terminal, Goose executes them directly. Its plugin architecture means you can extend the agent with custom tools without modifying the core codebase. MCP support lets Goose connect to databases, file systems, and APIs through standardized server configurations.

The agent maintains context across tool calls, so multi-step workflows (like setting up a new service, writing tests, and deploying) happen in a single conversation without re-explaining the project context.

How to use

1. Install Goose:

brew install block/tap/goose
# or
pipx install goose-ai

1. Start a Goose session:

goose session start

1. Describe what you want to accomplish. Goose will plan and execute the steps, asking for confirmation before destructive actions.

Example

# Start a Goose session and ask it to set up a project
goose session start

# In the Goose prompt:
> Create a new FastAPI project with a health check endpoint,
> a Dockerfile, and a GitHub Actions CI pipeline.
> Run the tests to make sure everything works.

Goose creates the project structure, writes the code, builds the Docker image, and runs the test suite. Each step is visible in the terminal with the option to approve or modify before execution.

Related on TokRepo

• AI agent tools -- Compare other AI agents for developer workflows
• Automation tools -- Browse tools for automating repetitive development tasks

Common pitfalls

• Goose executes real commands on your system. Always review the proposed actions before confirming, especially for destructive operations like file deletion or production deployments.
• Plugin configuration requires a YAML file. Missing or malformed configuration is a common source of startup errors. Validate your config before starting a session.
• Goose requires an LLM API key (OpenAI, Anthropic, or others). The agent itself is free, but you pay for the underlying model's token usage.