Creativity · MCP — client

MCP Goose Client (Block)

Goose is Block's open-source, extensible AI agent designed to run on a developer's machine. MCP is its extension mechanism: every capability (filesystem, GitHub, browser, databases) is an MCP server that Goose connects to. Goose ships both a CLI and a desktop UI, and its recipes let teams pin known-good extension sets for specific workflows.

MCP facts

Kind: client
Ecosystem: anthropic-mcp
Language: Rust + TypeScript
Transports: stdio, http

Capabilities

Runs MCP servers as 'extensions' with a single toml config
CLI (`goose session`) + desktop UI for chat-driven automation
Recipes: versioned combinations of extensions for repeatable workflows
Supports Anthropic, OpenAI, Google, and local Ollama providers

Install

curl -fsSL https://github.com/block/goose/releases/latest/download/download_cli.sh | sh

Configuration

# ~/.config/goose/config.yaml
providers:
  anthropic:
    api_key: ${ANTHROPIC_API_KEY}
extensions:
  - name: filesystem
    command: npx
    args: ["-y", "@modelcontextprotocol/server-filesystem", "~/projects"]
  - name: github
    command: npx
    args: ["-y", "@modelcontextprotocol/server-github"]
    env:
      GITHUB_TOKEN: ${GITHUB_TOKEN}

Frequently asked questions

How is Goose different from Claude Desktop?

Goose is open-source, multi-provider, and has first-class autonomous loops ('goose session'). Claude Desktop is Anthropic's polished consumer client. Both speak MCP.

What are Goose recipes?

A recipe bundles a prompt, extensions, and model settings into a YAML file. Teams share recipes to standardize 'triage this bug' or 'review this PR' workflows.

Does it support background execution?

Yes — Goose supports running an autonomous session with step-by-step approval or a looser 'autopilot' mode, depending on the recipe and configuration.

Sources

Goose — Block's open-source AI agent — accessed 2026-04-20
Goose GitHub — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20