Home / AI Tools / Goose
Goose

Goose

Open-source autonomous AI coding agent by Block

API costs only (varies by model) ✗ Split Opinion Visit Website ↗

Score Breakdown

7.8
7.8
9.1
Code Quality & Accuracy 8.0
7.5 7.7 8.8
Context Understanding 8.0
7.5 7.6 9.0
Multi-file Editing 8.2
7.5 7.8 9.2
Speed & Performance 8.0
7.0 7.5 9.5
Pricing Value 9.0
9.5 8.0 9.5
Ease of Use 7.3
6.5 7.0 8.5
Model Flexibility 9.6
9.5 9.2 10.0
Extension Ecosystem 8.1
8.0 8.4 8.0

Judge Opinions

Claude Opus 7.8

"Goose is Block's fully open-source (Apache 2.0) AI coding agent with 30K+ stars, 25+ model providers, and a unique Recipes system that enables reusable YAML workflow definitions for team-wide automated tasks. Block's internal adoption (60% of 12,000 employees using it weekly) and co-founding of the Linux Foundation Agentic AI Foundation provide unusual corporate credibility. The trade-off is that Goose is purely an orchestration layer — it adds no proprietary intelligence, and Block's own red team successfully exploited it via prompt injection in poisoned recipes."

+ Recipes system enables reusable YAML workflow definitions with goals, required extensions, and structured inputs — a genuinely unique differentiator
+ 25+ model providers including local models via Ollama and Docker, making it completely free to run with open-source models
+ Block corporate backing with 60% internal adoption across 12,000 employees, plus co-founding the Linux Foundation Agentic AI Foundation
+ First-class MCP support with 3,000+ tool integrations for GitHub, Slack, Jira, Docker, Kubernetes, and more
+ Extremely active development: 2-3 releases per week with Rust core providing performance and memory safety
- Purely an orchestration layer with no proprietary intelligence — output quality is entirely determined by the chosen model
- Block's red team ('Operation Pale Fire') successfully exploited the recipe system via invisible Unicode prompt injection characters
- Setup requires terminal comfort, API key configuration, and YAML management — described as 'fiddly' by users
- Desktop GUI feels like an afterthought compared to the CLI, and the MCP implementation is behind the latest June 2025 standard
GPT-5.2 7.8

"Goose is an open-source developer agent focused on actually doing work in your environment: reading files, editing code, and running commands as part of multi-step tasks. It’s model-agnostic and extensible (including MCP support), which makes it flexible for teams, but the setup and reliability depend heavily on your chosen model provider and permissions configuration."

+ Open-source agent designed around real repo workflows (files + commands + patch review)
+ Provider-agnostic configuration supports a wide range of model backends
+ Extensible via MCP and custom instructions for connecting additional tools
+ Works well for longer tasks through persistent sessions and repeatable recipes
+ Transparent development via public docs, releases, and issue tracking
- Requires more setup than IDE-native assistants (provider keys, permissions, and runtime config)
- End-to-end task speed can be slow on complex jobs because it executes many tool steps
- Quality varies significantly by model/provider choice, making results less predictable
- Needs disciplined permissioning and review to avoid unsafe command execution
Gemini 3 9.1

"Goose is a robust, open-source autonomous agent backed by Block (Square) that excels at multi-step engineering tasks. Written in Rust for performance, it offers a polished Desktop App and CLI that integrate seamlessly with Model Context Protocol (MCP) servers, making it a highly performant and flexible choice for developers who want 'off-the-shelf' autonomy without vendor lock-in."

+ Rust-based core delivers exceptional performance and stability
+ Model-agnostic design supports any LLM (local or cloud)
+ First-class support from Block ensures active enterprise-grade maintenance
+ Native Desktop App provides a friendlier entry point than pure CLI agents
- Requires bringing your own API keys which can be costly for autonomous loops
- Extension ecosystem is growing but smaller than VS Code-native tools
- Autonomous loops can occasionally spiral if not monitored

/// RECOMMENDED_USE_CASE

"Developers wanting a fully autonomous, model-agnostic coding agent with MCP integration for end-to-end workflows"