MCP Server vs REST API: What's the difference?
Why do we need a separate thing for LLMs when we already have REST?
Welcome to Infinite Curiosity, a newsletter that explores the intersection of Artificial Intelligence and Startups. Tech enthusiasts across 200 countries have been reading what I write. Subscribe to this newsletter for free to directly receive it in your inbox:
MCP server and REST API look very similar. Why do we need a separate thing for LLMs when we have REST API? Here's the SINGLE BIGGEST difference between the two:
Statefulness.
Okay what exactly does that mean?
REST APIs provide stateless endpoints. It means that each request is isolated, so clients must rehydrate context or chain resources manually. In other words, they won't remember anything that was previously asked.
Also you can ask the same question hundred times and it will provide the same response. In other words, it's idempotent. Just like how "multiply by 1" results in the same thing regardless of how many times you apply it to a number.
In MCP servers, context is a first-class citizen. It means that every call includes a thread-id. This preserves conversational state, tool invocations, and memory. It turns the server into a lightweight agent runtime.
REST excels at predictable micro-service boundaries and idempotent transactions.
MCP shines in multi-step reasoning loops e.g. summarize a PDF, spawn a tool, refine an answer (without glue code).
If you’re building LLM-native workflows where people expect statefulness, choose MCP server. If you're delivering web resources or integrations, you should use battle-tested REST API.
If you're a founder or an investor who has been thinking about this, I'd love to hear from you.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 friend who’s curious about AI: