Goose

Name: Goose
Author: AgentRadar

Verified

Block's MCP-native coding agent with extensible plugin architecture

Coding Agents Open Source

8.5 /10

Visit Website GitHub

Released

2024

Country

United States

API

N/A

Self-Host

Yes

GitHub Stars

50,000

Funding

Public (Block)

Last Updated

2026-06

About Goose

Goose is Block's (formerly Square) open-source coding agent, and it is one of the first agents built natively around the Model Context Protocol — MCP servers are not an add-on but the core extension model. This means any capability Goose needs — file editing, terminal access, database queries, browser control, internal APIs — comes from an MCP server that can be added, swapped, or custom-built. Goose runs as both a CLI tool and a desktop application, and supports Claude, GPT, Gemini, and local models via BYOK. It includes built-in developer tools for file editing and terminal execution, and adds scheduling for running tasks on a timeline and session management for resuming work across restarts. The MCP-native architecture makes Goose uniquely suited for automation that extends beyond pure coding — any tool that exposes an MCP server becomes a Goose capability. As one of the first agents to go all-in on MCP, it benefits from the rapidly growing ecosystem of community and vendor MCP servers.

Verdict

The MCP-native coding agent — if you believe MCP is the future of agent extensibility, Goose is the most committed implementation. Built by Block with enterprise credibility, it offers a plugin architecture that goes beyond coding into general automation. Less polished than Cursor or Claude Code, but unmatched as a foundation for teams building custom MCP-powered workflows.

Features

MCP-first architecture — all capabilities are MCP servers

CLI and desktop application modes

Bring-your-own-key for Claude, GPT, Gemini, local models

Built-in developer tools: file edit, terminal

Task scheduling for timed or recurring execution

Session management and resume

Extensible via custom MCP servers

Detailed Ratings

Ease of Use

8.0

Value for Money

9.0

Features

8.5

Support

8.0

Performance

8.5

Overall Rating

8.5 /10

Pros & Cons

Pros

MCP-native architecture — any MCP server becomes a capability, not just coding tools
Built by Block (Square) with engineering rigor and enterprise credibility
Model-agnostic BYOK with support for frontier and local models
Task scheduling turns the agent into an automation engine, not just a coding tool
Session resume lets you pick up where you left off across restarts

Cons

MCP-centric design has a learning curve — understanding and configuring MCP servers requires upfront investment
Less polished UX than Cursor or Copilot — closer to a developer tool than a consumer product
BYOK cost management is on the user — variable spend depending on model and usage

Use Cases

MCP-native automation across coding and non-coding toolsScheduled or recurring development tasks with task runnerTeams building custom MCP servers for proprietary workflowsDevelopment with model choice and local-model support for air-gapped environmentsSession-based work that needs pause and resume across days

Who Is It For?

Developers and teams who want an MCP-native agent with extensible plugin architecture for coding and broader automation tasks

#cli#desktop#mcp-native#open-source#extensible#automation

Pricing Plans

Open Source

Free

Full CLI and desktop application
Bring your own LLM API key
MCP server ecosystem access
Apache 2.0 license

LLM API Costs

Usage-based /BYOK

Pay your chosen LLM provider directly
MCP servers may have their own API costs
Local models via Ollama are free

Frequently Asked Questions

What does MCP-native mean?

Goose is built around the Model Context Protocol as its core architecture. Every capability — file editing, terminal access, browser control, database queries — is provided by an MCP server rather than being hardcoded. This means you add, remove, or swap capabilities by connecting different MCP servers, making Goose endlessly extensible.

Is Goose free?

Yes — Goose is Apache 2.0 licensed and fully free. You pay only for the LLM API keys you bring, and you can use local models via Ollama or similar for zero API cost.

How is Goose different from Claude Code or Cursor?

Goose is MCP-native — its extensibility comes from connecting MCP servers, which can do anything from file editing to database queries to browser automation. Claude Code is terminal-native with deep Claude integration. Cursor is an AI IDE. Goose's strength is the MCP plugin architecture; its limitation is that it requires understanding the MCP model to get the most out of it.