1

The MCP standard released in Nov 2024 HTTP+SSE was recently deprecated and the protocol change to "HTTP streaming". But looks like MCP now default to just old fashioned non streaming HTTP requests by a client with SSE is still there as an option, with the only difference that some json snippet defines the default request/response flow as "batch" which are HTTP request very similar to REST: GET, POST, DELETE, OPTIONS and a standard 'endpoint' (a fixed HTTP path). Blog/Vlogs posts seem to make you believe this as a revolutionary change that is more modern and also faster.

I don't see the innovation here and why this is called streaming why it per default seems not use it ? Is HTTP streaming anything new or a marketing gem ? It looks like the whole MCP "standard" wasn't thought through very well before releasing it seen the fast changes or it must be released to a certain project management / development methodology ?.

  [1]: https://www.claudemcp.com/docs/streamable-http
  [2]: https://blog.christianposta.com/the-updated-mcp-oauth-spec-is-a-mess/
  [3]: https://mcp-framework.com/docs/Transports/http-stream-transport/
  [4]: https://www.anthropic.com/news/model-context-protocol
1
  • 1
    A quick glance at the MCP standard suggests that their new 'Streamable HTTP' transport still does use text/event-stream, which is SSE. Commented Jun 9 at 17:24

1 Answer 1

0

You can make HTTP GET to <host>/mcp and accept incoming events. That's all about streaming.

https://modelcontextprotocol.io/specification/2025-03-26/basic/transports#listening-for-messages-from-the-server

Sign up to request clarification or add additional context in comments.

3 Comments

Digging a bit more can only conclude (so far) HTTP+SSE is the same except streaming HTTP uses one endpoint (/mcp) call that optionally creates a persistent SSE connection from server to client (and if not, is it just plain non persistent, non streaming, HTTP), whereas HTTP+SSE ALWAYS requires one endpoint for persistent data from the server (/sse) and another one (/messages) from client to the server. Some articles .. aibase.com/news/16393 make it look like data over the internet will suddenly be much faster.
HTTP+SSE ALWAYS requires one endpoint for persistent data from the server (/sse) - I am looking for an example of this setup; can you help?
I wish I had. AFAIK HTTP streaming was created to reduce the cost for serverless solutions, like Amazon. All Claude local run examples are stdio (stateful) use npm, python etc..which creates an server instance. I assumes a stateless HTTP server without SSE (i.e just a local Apache PHP server) based on HTTP streaming should be possible. can't find any client that uses HTTP-streaming w/o using the SSE option. Claude Desktop for Linux isn't there. The inspector tool supports stdio, HTTP+SSE and HTTP-streaming, but all with SSE. Why they did not create stateless HTTP MCP?

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.