Goose
Block’s open, on-machine agent — runs locally, any LLM, MCP-native.
Standout features
Goose runs on your machine and turns an LLM into real action — building, editing, running and testing through MCP tools.
Worldwide search interest, indexed 0–100 · Google Trends.
Goose (“codename goose”) is an open-source, on-machine AI agent created at Block — a general-purpose agent that turns LLM output into real action.
- Created by Block (ex-Square); launched January 2025.
- Runs locally and works with any LLM.
- MCP-native — capabilities come from the tools you plug in.
- Built in Rust; desktop app, CLI and API.
Goose is a blank slate defined by its tools.
- On-machine execution keeps code local by default.
- 15+ model providers, including local via Ollama.
- “Recipes” (YAML macros) capture repeatable workflows.
- Not just code — research, automation and data tasks too.
Free and open source — you supply the model.
- No license cost; BYO API key (or local model).
- Costs scale with whatever model you choose.
- Runs offline-capable with local models.
Goose fits developers who want control and extensibility.
- Privacy-minded teams wanting on-machine execution.
- Tinkerers who like composing tools via MCP.
- People who want one agent for code and non-code work.
- People who want top-tier coding quality out of the box.
- Beginners uncomfortable in the terminal.
No tool is perfect — the trade-offs to weigh:
- Quality tracks your model choice.
- Recipes have a learning curve.
- Terminal comfort helps a lot.
- Raw coding trails Claude Code and Cursor.
- ✓Open source and runs on your machine
- ✓Works with any LLM (15+ providers, local too)
- ✓MCP-native and highly extensible
- ✓Desktop app, CLI and API
- ✓Good fit for air-gapped/private setups
- ✕Quality depends heavily on your model
- ✕Recipes and setup have a learning curve
- ✕Terminal comfort recommended
- ✕Raw coding trails the leaders
Developers like that Goose runs locally, works with any model and bends to whatever MCP tools they bolt on — a genuinely flexible, private agent. The flip side they note is that output quality depends on the model and the recipes take some learning. Sentiment is positive among hands-on, control-minded developers.
Goose was created by Block (formerly Square) and is now stewarded as an open-source project.
Company figures are drawn from public disclosures and reputable trackers (gathered Jun 2026). User and revenue numbers are estimates and move fast.
Pick up to two other coding tools to see them head-to-head on the same rubric.