What is an MCP SSE (Server-Sent Events)?

In MCP's early network-transport design, a client would open a long-lived SSE connection to receive a stream of server-initiated messages such as tool results and notifications, while sending its own requests as separate HTTP POSTs that the server correlated back to the open stream. This let MCP work over standard HTTP infrastructure — load balancers, proxies, browsers — without needing a raw socket protocol, which mattered for remote, cloud-hosted servers that couldn't just run as a local stdio subprocess. In practice, the two-connection SSE design proved awkward: it complicated resumability after a dropped connection, sat poorly with some proxies and serverless hosting models, and required managing two separate HTTP flows for one logical session. The MCP specification has since superseded it with Streamable HTTP, a single-endpoint transport that can optionally upgrade to a stream but doesn't require the split-connection model. SSE-based MCP servers still exist in the wild and some clients retain backward compatibility with it, but new server implementations are generally steered toward Streamable HTTP rather than plain SSE.