MCP server or API?When you need both
Updated · 4 min read
If you already have a REST API, why build an MCP server? In short: an API is for developers, MCP is for AI assistants. Here is the difference, when an MCP server wraps your existing API, and when the API alone is enough.
The short answer
- An API (REST or GraphQL) is a contract between programs. A developer reads the docs and writes code that calls the endpoints.
- An MCP server is a layer on top of the API, built for AI models. The assistant fetches the list of tools, reads their descriptions and decides when to use them. Nobody writes code on the ChatGPT or Claude side.
- It isn't one or the other. An MCP server usually calls your existing API.
How they differ in practice
- Discovery: the AI app asks the MCP server for its tools and gets names, descriptions and parameter schemas. With an API, someone first has to read the docs and code the integration.
- Audience: an API returns data for code. An MCP tool returns an answer the model has to understand and pass on to the user.
- Reach: one MCP server works in ChatGPT, Claude, Cursor and Visual Studio Code. An API integration is separate code in every app.
- User consent: MCP tools are marked as read-only or as changing data, so apps know when to ask for permission.
- Sign-in: the MCP specification describes authorization on OAuth 2.1, so every AI app signs users in the same way. With an API, each integrator does it their own way.
Why not expose the whole API as tools
It's tempting to generate a tool from every endpoint, for example from an OpenAPI spec. But assistants do worse when there are many small tools: they have to pick the right one, and every description takes up room in their context.
It's better to design a few tools around what users want to get done. "Quote an order with delivery" instead of three separate calls for the product, the price list and the courier. Claude's directory rejects a single generic api_request tool that accepts any endpoint.
When an API alone is enough
- You connect two systems with no person involved, such as a store and an ERP or warehouse.
- Partners build their own apps on your data.
- You write your own agent in code that calls specific functions. MCP is optional here, but it helps when the same set of tools has to work in several apps.
When you need an MCP server
- You want customers to find and buy your products in ChatGPT or Claude. We cover this in How to sell in ChatGPT.
- Your users work in Claude or ChatGPT and want your product's data there. Details in Claude connector for SaaS.
- You want a listing in the ChatGPT Plugin Directory or Claude's Connectors Directory. Both accept MCP servers, not plain APIs.
- Your team wants to work with company data from an assistant: reports, order handling, searching documents.
What it looks like technically
An MCP server is usually a small service next to the API, on your domain (for example at /mcp), reachable over the Streamable HTTP transport. Each tool has a name, a description, a JSON Schema for its parameters and a function that calls the API and turns the result into a readable answer.
Permissions stay in the API. The server acts on behalf of the signed-in user (with its own token for the API, not the one from ChatGPT or Claude), so the assistant sees exactly what that user sees in your app. That's how we built the server for Drukalo: search, quotes and the cart use the same data as the store.
- Does MCP replace a REST API?
- No. The MCP server uses the API and adds a layer for AI assistants. Your integrations, mobile apps and partners keep using the API.
- Can I generate an MCP server from an OpenAPI spec?
- Yes, there are tools that do it, and it's a good starting point for testing. Before you submit to the directories, review the list and merge small tools into a few built around specific tasks.
- Does an MCP server add security risk?
- It exposes only what you program, and it runs with the signed-in user's permissions. The basics: separate read and write tools, OAuth instead of keys pasted into a chat, and the same rate limits as your API.
- MCP integration: your ERP, store and CRM in ChatGPT and Claude
- Aesthetic clinic websites in Poland
- Online booking for a private clinic
- An AI assistant for your clinic: what's allowed in 2026
- ChatGPT product feed: get your products into ChatGPT shopping
- Clinic advertising in Poland: what a website may say
- Your clinic in ChatGPT and Google
- Patients from abroad: a medical tourism website
- Claude connectors: what they are and how to add one
- MCP servers in ChatGPT: how to connect one
- How to connect WooCommerce to ChatGPT in 2026
- OAuth for MCP servers: sign-in in ChatGPT and Claude
- MCP Apps: interface in ChatGPT and Claude
- MCP server security: what to check before launch
- How to test an MCP server in ChatGPT and Claude
- Hiring a software studio in Poland
- Your Shopify or WooCommerce store in ChatGPT
- What is an MCP server? Explained for online stores
- Claude connector for SaaS: what it takes
- How to sell in ChatGPT in 2026
- MCP server, ChatGPT plugin or Claude connector?
- How to get listed in ChatGPT and Claude
Have an API and want an MCP server?
MCP servers