MCP Servers

MCP (Model Context Protocol) servers extend the CLI's capabilities by providing additional context and tools to AI models.

What are MCP Servers?

MCP servers:

Provide custom tools and resources to the CLI
Enable integration with external services
Expose domain-specific knowledge and utilities
Allow AI models to interact with specialized systems

Server Types

Tool Servers

Provide callable tools:

mcp_servers:
  - name: my-tools
    type: tool
    handler: ./servers/my-tools.ts

Resource Servers

Expose accessible resources:

mcp_servers:
  - name: my-resources
    type: resource
    handler: ./servers/my-resources.ts

Building an MCP Server

// servers/my-server.ts
export const createServer = () => {
  return {
    name: 'my-server',
    tools: [
      {
        name: 'my-tool',
        description: 'What this tool does',
        execute: async (args) => { /* ... */ }
      }
    ]
  }
}

Configuration

Configure servers in your CLI settings:

mcp_servers:
  - name: my-server
    config:
      apiKey: ${SOME_API_KEY}
      timeout: 30000

Learn about skills for CLI extensions
Explore hooks for execution lifecycle control
See best practices for optimization