Skip to content

Quick Start

Create a config file

Create mcpbox.json in your working directory:

json
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    },
    "sequential-thinking": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
    }
  }
}

Run

bash
npx mcpbox

MCP server commands (e.g., uvx, docker) must be available where MCPBox runs.

Or with Docker:

bash
docker run -v ./mcpbox.json:/config/config.json -p 8080:8080 ghcr.io/kandobyte/mcpbox

The Docker image includes Node.js and Python, supporting MCP servers launched via npx and uvx.

Connect a client

MCPBox starts on http://localhost:8080. Add it to your MCP client's config:

json
{
  "mcpServers": {
    "mcpbox": {
      "url": "http://localhost:8080"
    }
  }
}

Next steps

  • Configuration — configure logging, server options, tool filtering
  • Authentication — configure API key or OAuth
  • Deployment — expose remotely with cloudflared or a reverse proxy
  • Connect AI — Claude Web, Claude Code, and other clients