How to Turn Any REST API into an MCP Server (2026 Guide)

The pattern is clear: developers don't just want to build MCP servers from scratch - they want to wrap the APIs they already have. "API to MCP" landed at #2 on Product Hunt on June 19, 2026 with 114 upvotes, and the Skillselion catalog's 7,730 MCP listings are full of exactly this pattern: Stripe MCP, GitHub MCP, database MCPs, internal-API MCPs. This guide explains the conversion workflow and how Claude Code handles it.

Key takeaways

• "API to MCP" was #2 on Product Hunt June 19, 2026 (114 upvotes) - the leading indicator of developer demand
• The Skillselion catalog tracks 7,730 MCP listings - the fastest-growing segment across 82,301 total listings
• The conversion pattern: REST endpoint → MCP tool definition → tool handler → server registration
• Claude Code handles the boilerplate; the key decisions are authentication, tool-name semantics, and response shaping
• For building MCP servers from scratch, see the companion MCP server build guide

Why convert an existing API to MCP instead of building from scratch?

MCP (Model Context Protocol) is the standard for giving AI agents structured access to external data and actions. If you already have a REST API - internal tooling, a product API, a third-party integration - wrapping it as an MCP server means any MCP-compatible agent (Claude Code, Cursor, other Claude SDK integrations) can use it immediately, without per-agent glue code. The 7,730 MCP listings in the Skillselion catalog (across 82,301 listings) represent exactly this: teams wrapping their own APIs once and making them available to all their agents.

What is the anatomy of an MCP tool definition?

Every REST endpoint maps to an MCP tool via three components: 1. Name - a clear, verb-noun identifier (get_user, create_invoice, search_products) that the LLM can understand and call correctly 2. Description - a sentence that tells the LLM when to use this tool (not just what it does) 3. Input schema - a JSON Schema object that validates the tool's parameters before the handler fires

The MCP SDK (TypeScript or Python) generates the server boilerplate; Claude Code with the MCP skills handles the translation from your OpenAPI spec or API docs.

How do I convert an OpenAPI spec to MCP tools with Claude Code?

The workflow Claude Code follows with the right skills: 1. Parse the OpenAPI YAML/JSON spec to identify endpoint groups 2. Group related endpoints into logical MCP tools (not necessarily 1:1 - one MCP tool can aggregate multiple REST calls) 3. Generate MCP tool definitions with name, description, and JSON Schema input 4. Write the handler that calls your REST API with the correct auth headers 5. Register the tools on the MCP server and expose the server's stdio or SSE transport

For the full MCP server authoring workflow, install the skills from the MCP server build guide and the best MCP server skills guide.

What are the key decisions when wrapping an API as MCP?

Authentication: How does the MCP server authenticate with your REST API? Common patterns:

• Environment variable secrets - simplest; the server reads process.env.API_KEY and injects it into every request
• OAuth flow - for user-scoped APIs; the MCP server handles the OAuth dance and passes the token
• mTLS - for internal APIs behind a service mesh

Tool granularity: One MCP tool per REST endpoint produces a noisy tool list that confuses the LLM. Group by user intent ("what does the agent want to accomplish?"), not by HTTP verb. A single manage_user tool that handles create/update/fetch is often better than three separate tools.

Response shaping: MCP tools return text or structured JSON. Shape responses to include only what the LLM needs - verbose API responses with metadata, pagination tokens, and internal IDs bloat the context and reduce reasoning quality.

How many MCP servers are in the Skillselion catalog?

As of June 2026, the Skillselion catalog tracks 7,730 MCP server listings across 82,301 total listings, with 97.8M total installs across all skill types. The MCP segment is the fastest-growing category - a direct reflection of the "wrap your API" pattern this guide covers. Browse them at the best MCP server skills guide.

A complete API-to-MCP conversion workflow

1. Document your API - if no OpenAPI spec exists, generate one from your existing routes 2. Group endpoints into logical MCP tools by user intent (not by HTTP method) 3. Scaffold the MCP server using the MCP TypeScript or Python SDK 4. Write tool handlers - each handler calls your REST API with proper auth 5. Test with Claude Code - install your local MCP server in Claude Code and verify tool descriptions are clear enough for the LLM to call them correctly 6. Reference the MCP server build guide for the server scaffolding patterns

Common pitfalls

• 1:1 endpoint-to-tool mapping - 50 REST endpoints should not become 50 MCP tools; group by user intent or the LLM will choose the wrong tool consistently
• Missing error messages in tool responses - when your REST API returns a 4xx, the MCP tool should return a human-readable error message, not the raw HTTP status code; the LLM needs to understand the error to decide what to do next
• Skipping the tool description - the tool name alone is insufficient; the description tells the LLM when to use the tool vs alternatives; a blank description breaks LLM tool selection

The full MCP ecosystem - 7,730 servers in the Skillselion catalog - shows what's possible. The best MCP server skills guide ranks the most-installed MCP skills by category.

Common questions