AgentBrew serves as a comprehensive Model Context Protocol (MCP) multiplexer, designed to streamline the integration of AI tools. This centralized platform enables AI developers to configure their tools just once, making them accessible across various AI agents such as Claude Code, Gemini CLI, and Cursor.

Core Concept

The challenge of fragmentation in AI development is prevalent, as each agent typically manages its MCP servers and tools in unique ways. AgentBrew addresses this issue by functioning as a Universal "USB Hub" for MCP, allowing the installation of tools to be utilized universally across all agents. Switching between agents is hassle-free, as users carry their full suite of tools and skills seamlessly.

Key Features

• Lazy Loading: MCP servers are initiated only when a tool is invoked, conserving resources.
• Auto-Discovery: Automatically identifies MCP servers in Node.js, Python, and Markdown projects, simplifying setup.
• Universal Migration: Import existing configurations from Gemini, Claude Code, and Cursor effortlessly.
• Instruction Index: Automatically makes GEMINI.md and CLAUDE.md readily available as resources for agents.

Usage Overview

To connect tools in AgentBrew, users can utilize commands such as:

# Install a tool from a Git URL
agentbrew install <github-url>

# Migrate from Gemini, Claude, or Cursor
agentbrew migrate

Tool management is equally intuitive, enabling command execution to list, enable, or update packages:

# List all tools and resources
agentbrew list

# Enable or disable a package
agentbrew enable <package-name>

# Uninstall a package
agentbrew uninstall <package-name>

Connecting Agents

AI agents can be configured to point to AgentBrew as the MCP server.

For example:

• Gemini CLI: gemini mcp add agentbrew agentbrew
• Claude Code: /plugin add agentbrew agentbrew
• Cursor: Configure a new command type MCP server with agentbrew.

In agents requiring manual JSON configuration, users can add AgentBrew to the MCP settings:

{
  "mcpServers": {
    "agentbrew": {
      "command": "agentbrew"
    }
  }
}

Architecture

AgentBrew employs an mcp-manifest.json cache within each package directory, facilitating rapid startup. The Router serves as a dynamic proxy that spawns child MCP processes on demand and routes requests efficiently using a prefix__name convention.