Day 4 — Goose: The MCP-Native Agent Harness

Goose began at Block (Square) as an internal AI agent. In 2026, Block donated it to the Linux Foundation's newly formed Agentic AI Foundation (AAIF), alongside Anthropic's Model Context Protocol (MCP) and the Agents.MD specification. Goose is now the reference implementation for MCP-native agents: local-first, model-agnostic, and extensible through 70+ MCP server connections. This day covers the tool that treats every external system as an MCP endpoint.

Lessons

1. What Goose Is & The AAIF Story — the Block-to-Linux Foundation transition, the AAIF mission, and where Goose fits in the ecosystem
2. Installation & First Run — installing Goose, connecting a model, and running your first task
3. MCP Extensions — connecting to 70+ MCP servers for infrastructure, data, and tool control
4. Local-First Architecture — how Goose runs locally, privacy model, and offline capabilities
5. Custom Extensions & Automation — building your own MCP servers for Goose and automating workflows

Next: Capstone: Multi-Tool Dev Workflow