Anthropic Significantly Updates MCP

Anthropic Significantly Updates MCP

The update has been called the largest since its launch. Let's take a look at what has changed.

Previously, MCP operated as a bidirectional connection with session state retention. This meant that when connected, the server opened a long-lived connection, and all context between requests lived in the memory of a single running process. This required a constantly running instance that hung around waiting, existing all the time while the session was active.

In the new version, MCP has become stateless, meaning that now each request is self-contained. The client sends all the necessary context in the request (or it is stored externally), and the server simply receives the request, processes it, returns the response, and forgets everything.

All serverless platforms (AWS Lambda, Cloudflare Workers, etc.) operate this way. Before this version, MCP could not be properly run on such infrastructure, but now it can: the developer's server can consist simply of a function that wakes up on request and goes to sleep.

In short: MCP has just become much cheaper + it has become much easier to scale. Practically, this means that the barrier to entry for creating an MCP server has dropped sharply, and we can expect explosive growth in the number of connectors for a wide variety of applications, even the smallest ones.