Goose ​

Goose is a configurable AI agent from Block, Inc. designed for software engineering. It runs locally on your machine, supports swappable LLM providers, and uses MCP as the backbone of its extension system — every Goose extension is, in fact, an MCP server under the hood.

Go
Goose
Block, Inc. · github.com/block/goose
TRANSPORT
Streamable HTTP ✓
PLATFORM
Windows · macOS · Linux
MCP VIA
CLI + Profiles

Extensions Are MCP Servers ​

Goose's architecture treats every extension as an MCP server. This means adding Vinkius Cloud gives the agent the same first-class access as its built-in tools for file editing, terminal commands, and web browsing.

Profiles let you bundle specific MCP servers and model settings into reusable presets. For example, a "backend-dev" profile might include a database MCP server and a monitoring tool, while a "frontend" profile loads a design-token API. Switching between them is a single command.

Other features that complement MCP:

  • Session persistence — Goose remembers tool results across restarts
  • Provider flexibility — Claude, GPT, Gemini, Ollama, or any OpenAI-compatible endpoint
  • Built-in safety — approve or deny tool calls before execution
  • 33k+ GitHub stars — active open-source community

Connecting Vinkius Cloud ​

1. Create a Connection Token ​

Log in to Vinkius Cloud, select your server, and generate a token. The URL format is:

https://edge.vinkius.com/{TOKEN}/mcp

2. Register via Goose CLI ​

bash
goose configure

Select Add Remote Extension and paste your Vinkius URL. Goose writes the config and validates the connection in one step.

Alternatively, add to your ~/.config/goose/profiles.yaml:

yaml
default:
  extensions:
    vinkius:
      type: remote
      uri: https://edge.vinkius.com/{TOKEN}/mcp

3. Start a Session ​

bash
goose session

Goose auto-discovers the available tools and uses them whenever they match the task at hand.

FAQ ​

Why does Goose use MCP for extensions? MCP provides a standardized protocol for tool communication. By adopting it, Goose ensures every extension works with the same transport, auth, and discovery mechanisms — no custom plugin APIs needed.

Can I approve MCP tool calls before they run? Yes. Goose has a built-in approval prompt for destructive actions. You can configure auto-approve for read-only tools while requiring confirmation for writes.

How do profiles relate to MCP? Profiles are named configurations that bundle an LLM provider, system prompt, and a set of MCP extensions. You can switch profiles mid-session with goose use <profile>.

Is Goose free? Open-source, Apache 2.0. Bring your own API key for the LLM provider of your choice.