Goose
Open-source AI coding agent from Block that runs locally on your machine. Supports any LLM provider, extends via MCP, and keeps your code private by default.
Overview
Goose is an open-source AI agent built by Block (Jack Dorsey's fintech company) that runs entirely on your local machine. Unlike cloud-hosted coding assistants, Goose operates as a standalone agent — available as both a CLI tool and a desktop app — that can autonomously navigate your filesystem, run shell commands, edit files, and execute multi-step development workflows. It's model-agnostic, meaning you bring your own API key from OpenAI, Anthropic, Google, Ollama, or any OpenAI-compatible provider and swap models freely.
What sets Goose apart from the growing crowd of AI coding tools is its extension model built on the Model Context Protocol (MCP). Rather than baking integrations into the core, Goose uses MCP servers as modular extensions — there's a built-in extension directory covering GitHub, Jira, databases, web scraping, and more. You can also write your own MCP extensions or point Goose at any existing MCP server. This makes it genuinely extensible beyond just code editing into research, DevOps, data analysis, and workflow automation.
With 45,000+ GitHub stars and an Apache 2.0 license, Goose has attracted a significant community. The tradeoff is clear: you get full privacy and zero subscription costs, but you're responsible for your own LLM API spend, and the setup requires more technical comfort than turnkey tools like Cursor or Windsurf. It's best suited for developers who want an autonomous agent they fully control.
Key features
Local AI Agent
Runs entirely on your machine as a CLI or desktop app. Reads and edits files, executes shell commands, and performs multi-step tasks autonomously without sending your code to a third-party service.
MCP Extensions
Extends capabilities through the Model Context Protocol. A built-in extension directory includes GitHub, Jira, databases, and web tools — or connect any custom MCP server for domain-specific workflows.
Multi-Provider
Model-agnostic architecture supports OpenAI, Anthropic, Google, Ollama, and any OpenAI-compatible endpoint. Swap models per session without reconfiguring the agent.
Pricing
Free tier: Completely free and open-source (Apache 2.0). You pay only for LLM API usage from your chosen provider.
| Plan | Price | What's included |
|---|---|---|
| Open Source | Free | Full agent capabilities, CLI + desktop app, all extensions, unlimited usage — bring your own LLM API key |
Full agent capabilities, CLI + desktop app, all extensions, unlimited usage — bring your own LLM API key
Pros & cons
Pros
- ✓Completely free and open-source with no vendor lock-in
- ✓Full local execution keeps your code private by default
- ✓MCP-based extension system is genuinely modular and community-driven
- ✓Model-agnostic — use any LLM provider or run local models via Ollama
Cons
- ×Requires your own API keys and managing LLM costs separately
- ×More setup friction than turnkey editors like Cursor or Windsurf
- ×No built-in IDE — it's an agent, not an editor, so you still need your own code editor
- ×Agent autonomy can be unpredictable without careful prompt guidance
