AAIF / Block · 2026-04-10 · major
Goose v1.2 — Automatic MCP Server Discovery, ACP Protocol, 44k GitHub Stars
Block's open-source AI agent hits v1.2 with automatic MCP server discovery, adoption of the Agent-Client Protocol (ACP) for agent-to-agent communication, and MCP Apps (interactive UI components from servers). Now under the AAIF foundation with 44k stars.
Goose adds MCP auto-discovery and ACP agent-to-agent protocol — the reference implementation for the evolving MCP ecosystem.
Key specs
| GitHub stars | 43,965 |
|---|---|
| Mcp extensions | 70 |
What is it?
Goose is an open-source, extensible AI agent (Apache 2.0, Rust) originally from Block that goes beyond code suggestions to install, execute, edit, and test with any LLM. Now steered by the AAIF foundation, it has become the de-facto reference implementation for Model Context Protocol extensions with 44k GitHub stars.
How does it work?
Version 1.2 adds automatic MCP server discovery, so Goose detects installed MCP servers without manual configuration. The Agent-Client Protocol (ACP) integration lets Goose act as a client to other agents — not just tools — analogous to how MCP standardized tool connections. MCP Apps support (formerly mcp-ui) enables servers to deliver interactive React-based UI components directly into the Goose interface.
Why does it matter?
Goose serves as the proving ground for new MCP features before they land in the official spec. The ACP addition signals the next wave of the protocol: agent-to-agent coordination beyond single-agent tool use. 70+ documented extensions make it the most MCP-integrated agent available.
Who is it for?
Developers who want a local-first, provider-agnostic AI agent with deep MCP integration and an active open-source community.
Try it
brew install goose-ai && goose run