goose
Open-source, local AI agent from Block that runs as a desktop app or CLI, works with 15+ model providers and extends through MCP.
Evaluation
What to know
Editorial notes based on the vendor's own documentation and public pricing, plus hands-on testing where a score or review is shown above. They describe fit for a job, not absolute quality.
Best for
- Developers who want an agent not tied to one model vendor
- Teams standardizing automations as shareable recipes in CI
- Power users connecting many MCP tools to one agent
Not ideal for
- Beginners who want a guided, fully managed coding product
- Organizations barred from letting agents run commands locally
- Users without access to a capable model API or local GPU
Key features
- Desktop app, CLI and API from one Rust codebase
- Works with 15+ LLM providers or existing subscriptions via ACP
- 70+ documented MCP extensions and interactive MCP Apps
- Shareable YAML recipes and parallel subagents
- Permission controls, sandbox mode and prompt-injection detection
Pros
- Completely free and vendor-neutral under Linux Foundation governance
- Runs locally, so files and tools stay on your machine
- Recipes make agent workflows repeatable across a team
Cons
- You still pay for whichever model provider powers it
- Autonomous actions on a local machine require careful permission setup
- Less hand-holding than commercial IDE-integrated agents
ToolifyNav verdict
goose stands out among coding and automation agents for being genuinely open and vendor-neutral, with foundation governance, broad provider support and one of the deepest MCP integrations available. Its recipes and subagents make it more than a chat-in-terminal tool, and the desktop app lowers the barrier for non-terminal users. Expect to spend time on permissions and model choice, and remember the agent is only as capable as the model behind it. For developers who prefer to own their tooling rather than rent an IDE plugin, it is an easy one to keep installed.
About
About goose
goose is a general-purpose AI agent that runs on your own machine. Originally released by Block and now part of the Agentic AI Foundation at the Linux Foundation, it is Apache 2.0 licensed and built in Rust. It ships as a native desktop app for macOS, Linux and Windows, a full command-line interface, and an API for embedding the agent elsewhere. Although many people adopt it for software work such as editing code, running tests and debugging, the project describes it as useful for research, writing, automation and data analysis too.
goose is model-agnostic: it supports more than 15 providers including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure and Bedrock, and can use existing Claude, ChatGPT or Gemini subscriptions through the Agent Client Protocol. It was an early adopter of the Model Context Protocol and documents more than 70 MCP extensions. Recipes capture workflows as portable YAML files that can run in CI, subagents handle parallel tasks, and MCP Apps can render interactive UIs inside the desktop app.
Security features include prompt-injection detection, tool permission controls, a sandbox mode and an adversary reviewer. The repository lists more than 55,000 GitHub stars and over 600 contributors.
Related
More in Coding & Development
Same category — not a curated competitor list.
Discover More
Weekly digest
Get weekly AI tools and hands-on reviews
A short digest of new tools and hands-on breakdowns. Unsubscribe anytime.