qvib.pro
RU

Goose

What for: a local open-source agent from Block — it runs engineering tasks on your machine and extends through MCP servers.

open-source бесплатно локально MCP профи

★46.1k GitHub checked 2026-06-01

Updated: 02.07.2026

$ curl -fsSL https://github.com/block/goose/releases/download/stable/download_c…

Open source ↗

Goose

What it is and who it's for

Goose is a local open-source AI agent from Block that carries out engineering tasks on your machine and extends through MCP servers (extensions). It's for developers and automation folks who want a hands-on agent running locally with any model and pluggable tools. What sets it apart is being MCP-native (huge extensibility) and shipping both a CLI and a desktop app.

Key features

  • A local agent: file edits, command execution, multi-step tasks.
  • MCP-native: plug in tools and services as extensions.
  • Any model: cloud providers or local ones (Ollama).
  • CLI and desktop app.
  • Recipes/automations for repetitive tasks.
  • Open source and free.

Get started in 5 minutes

  1. Install the CLI (or download the desktop app):
curl -fsSL https://github.com/block/goose/releases/download/stable/download_cli.sh | bash
  1. Configure your model provider and key:
goose configure
  1. Start a session and give it a task:
goose session

When to use it, and when not to

  • ✅ Use it if: you want a local hands-on agent with rich MCP extensibility and any model you like.
  • ✅ Use it if: you want to automate engineering tasks with recipes.
  • ❌ Skip it → go with "Cline"/"Kilo Code" if you work mostly inside VS Code.
  • ❌ Skip it → go with "Claude Code" if you want a ready-made pipeline with verification and subagents on the Anthropic stack.

The honest price

Free and open source; you pay for tokens from the model you choose (local models are free). Exact figures are with the provider and change over time.

Gotchas

  • MCP power means configuring the right extensions for the job.
  • Running commands locally — be careful with permissions and destructive actions.
  • Quality equals model quality; with a weak local model the agent flounders.
  • The docs are still evolving — sometimes you have to look in the repo.

🤖 Prompt accelerator

"I'm setting up Goose (CLI/desktop) for local task automation on . Help me pick a model that fits my budget, choose MCP extensions for the task (file work, databases, APIs, for example), and write a recipe for a repeating scenario. Give me the install/config commands and an example first session."

Читать по-русски →