# Standalone MCP Server Generation

Model Context Protocol (MCP) servers enable AI agents to interact with your APIs in a natural, conversational way. With Speakeasy, you can generate production-ready MCP servers directly from your OpenAPI/Swagger specification, without needing to write any code.

## What's New?

Previously, MCP servers were generated as part of TypeScript SDK generation using the `enableMCPServer` flag. That flag is now deprecated and should be left as `false`. Instead, you can generate standalone MCP servers independently:

- **Direct Generation**: Generate MCP servers directly from your OpenAPI spec
- **Language Agnostic**: Works with any API, regardless of the implementation language
- **Cloudflare Ready**: Optional Cloudflare Worker configuration for easy deployment
- **Desktop Extension Support**: Automatic generation of Anthropic Desktop Extensions (.dxt)

## Key Features

- **Automatic Tool Generation**: Each API operation becomes a discoverable tool
- **Type Safety**: Uses OpenAPI/Swagger types for accurate request/response handling
- **Customization Options**: Control tool names, descriptions, and more via OpenAPI extensions
- **Scoping Support**: Tag and control which operations are available to AI agents
- **Production Ready**: Built for reliability and performance at scale

## Getting Started

Follow our [step-by-step guide](/docs/standalone-mcp/build-server) to generate your first MCP server, or explore our guides on:

- [Customizing your MCP Server](/docs/standalone-mcp/customize-tools)
- [Deploying to Cloudflare Workers](/docs/standalone-mcp/cloudflare-deployment)

<Callout type="info">
  Want to understand MCP better? Check out our [introduction to MCP](/mcp/overview) and [AI agents overview](/mcp/using-mcp/ai-agents/introduction).
</Callout>