[MCP] add support for SSE + HTTP

[Wauplin]

Followed implementation from huggingface/huggingface.js#1422

cc @evalstate if you can have a look at it

How to test

import os

from huggingface_hub import ChatCompletionInputMessage, ChatCompletionStreamOutput, MCPClient


async def main():
    async with MCPClient(
        provider="nebius",
        model="Qwen/Qwen2.5-72B-Instruct",
        api_key=os.environ["HF_TOKEN"],
    ) as client:
        await client.add_mcp_server(type="sse", url="https://evalstate-flux1-schnell.hf.space/gradio_api/mcp/sse")

        messages = [
            {
                "role": "user",
                "content": "Generate a picture of a cat on the moon",
            }
        ]

        async for chunk in client.process_single_turn_with_tools(messages):
            # Log messages
            if isinstance(chunk, ChatCompletionStreamOutput):
                delta = chunk.choices[0].delta
                if delta.content:
                    print(delta.content, end="")

            # Or tool calls
            elif isinstance(chunk, ChatCompletionInputMessage):
                print(
                    f"\nCalled tool '{chunk.name}'. Result: '{chunk.content if len(chunk.content) < 1000 else chunk.content[:1000] + '...'}'"
                )


if __name__ == "__main__":
    import asyncio

    asyncio.run(main())

Output

Called tool 'evalstate_flux1_schnell_infer'. Result: '[Binary Content: Image image/webp, 38474 bytes]
The task is complete and the content accessible to the User
Image URL: https://evalstate-flux1-schnell.hf.space/gradio_api/file=/tmp/gradio/c178b5c1aae663d896b21cf6d0090e8e868e8a87e2d89b272243a9f135faf3ad/image.webp
42'

[hanouticelina]

Very nice and clean! thank you

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[julien-c]

Suggested change

	async def add_mcp_server(self, type: Literal["streamablehttp"], **params: Unpack[StreamableHTTPParameters_T]): ...
	async def add_mcp_server(self, type: Literal["http"], **params: Unpack[StreamableHTTPParameters_T]): ...

in the JS implem we've just used "http", and i think i've seen in some clients (Copilot config for instance) they went for just "http" too.

As it's expected to become the main method, i would keep it super short and simple

[julien-c]

(i meant VS Code not Copilot)

here, see this line: https://gist.github.com/AArnott/ea481eeb07fb78b4b6bf4b7436d88009#file-mcp-schema-json-L188

from discussion: modelcontextprotocol/modelcontextprotocol#292 shared by @evalstate

[Wauplin]

Thanks for noticing. Indeed much better to have "http". Addressed in 25726d8

[evalstate]

I was in the air while this happened (saw the email) @Wauplin don't think you can make that any shorter!