AgentZoo

Personal agent hub with MCP and a cozy UI. One server (HTTP API + MCP), one webapp, shared types.

Prerequisites

• Node.js 20+

• pnpm

Related MCP server: @willpowell8/cursor-cloud-agent-mcp

Setup

pnpm install

Scripts

Project structure

• packages/types — Shared TypeScript types (Agent, StoreShape)

• packages/server — Fastify HTTP API + MCP server (stdio), JSON file store

• packages/webapp — Vite + React + Tailwind, design tokens, Framer Motion

Docs

• TECH-STACK.md — Stack and tooling

• FRONTEND-LAYOUT.md — UI structure and flow

• AESTHETICS.md — Design tokens

• MCP-ARCHITECTURE.md — Store, server, and MCP flow

Cursor MCP

To use AgentZoo as an MCP server in Cursor, add to .cursor/mcp.json or Cursor settings:

{
  "mcpServers": {
    "agent-zoo": {
      "command": "pnpm",
      "args": ["--filter", "@agent-zoo/server", "run", "dev"],
      "env": {
        "AGENT_ZOO_STORE_PATH": "%USERPROFILE%/.agent-zoo/agents.json"
      }
    }
  }
}

On macOS/Linux use $HOME/.agent-zoo/agents.json for AGENT_ZOO_STORE_PATH.

Inject an agent via slash prompt

Once the MCP server is running, open a new chat in Cursor and type / to see MCP prompts. Select one of the AgentZoo prompts:

• Use Current Agent — injects the currently selected agent

• Use Agent: {name} — injects a specific agent

These prompts insert a user message that tells the assistant to adopt the agent's persona and active skills for the conversation.