mcp-com-ai/api-to-mcp-skills

Overview

This skill converts OpenAPI v3 specifications into fully functional MCP (Model Context Protocol) servers using the HAPI CLI. It requires no custom code and supports local development, Docker-based self-hosting, and Cloudflare Workers deployment. Use it to expose existing API functionality as MCP tools or to deploy API-backed services across multiple hosting targets.

How this skill works

The tool validates an OpenAPI v3 spec and generates an MCP server wrapper that proxies calls to your backend API. In headless mode it maps OpenAPI operations to MCP tools and exposes a Streamable HTTP transport at /mcp for JSON-RPC and streaming responses. You can run the generated server locally, inside Docker, or deploy it to Cloudflare Workers via built-in CLI commands.

When to use it

• You have an OpenAPI v3 spec and want MCP-compatible tool endpoints with zero code.
• You need to test or demo API-backed MCP tools locally before deployment.
• You want to deploy API-based MCP servers to Cloudflare Workers for low-friction hosting.
• You prefer a Docker-based self-hosted MCP server for production control.
• You need streaming responses (SSE) or JSON-RPC over HTTP for model integrations.

Best practices

• Confirm your OpenAPI spec is v3.x and resolves any $ref or schema errors before conversion.
• Use --headless when proxying an existing backend to avoid serving the docs UI.
• Validate /health and /mcp/ping after startup to ensure the service is healthy.
• Test at least one tool call through the /mcp transport to verify end-to-end behavior.
• Include clear error messages in your backend so the MCP wrapper can surface actionable errors.

Example use cases

• Wrap a public API (e.g., Petstore) as MCP tools for model-driven workflows and demos.
• Rapidly expose internal microservices as MCP tools for team experimentation.
• Deploy an API facade to Cloudflare Workers to minimize latency and maintenance.
• Run a self-hosted Docker MCP server for staging or production with controlled networking.
• Stream long-running operation outputs via SSE through the /mcp endpoint for real-time models.