can we call tools and return steaming output?

such as
@mcp.tool()
async def function(text: str):

url =URL

headers = {
    "User-Agent": USER_AGENT
}

async with httpx.AsyncClient() as client:
    try:
        async with client.stream("POST", url, headers=headers, json={"text": text}, timeout=100) as response:
            response.raise_for_status()
            async for chunk in response.aiter_bytes():
                yield chunk
    except Exception as e:
        print("error", e)

response
<async_generator object func at 0x107641ec0>

just return this

need this feature

Raise this Topic.

A Tool without Streaming Response makes just no sense!

This seems like a very important feature since it's a really common use case of tools.

As far as I know this isn't supported by the model context protocol itself. It's in the roadmap, and there's an open issue at that level (modelcontextprotocol/modelcontextprotocol#117). The best option right now for this is progress reporting, e.g.

for progress_percent in range(0, 100, 5):
            await ctx.report_progress(progress=progress_percent, 
                      total=100, message=f"Generating, {progress_percent}% done...")

More details on potential API in the Python SDK #784

May I ask if this issue has already been resolved? Is there a planned release date? Thank you.

Thanks for all the interest here. Streaming tool results isn't currently supported by the MCP spec - the Streamable HTTP transport's SSE response is for delivering a single JSON-RPC response plus server-initiated messages, not partial content.

What exists today for long-running scenarios:

Progress notifications (notifications/progress) - informational status updates, no partial results
Elicitation/Sampling - server can request input from client mid-tool-call via SSE
Tasks (experimental) - status polling for long-running operations, but the final result is still atomic

None of these support streaming partial tool content like LLM token streaming.

Active spec proposals:

Issue #117 - Original streaming tool results issue
PR #776 - Partial results with _meta.hasMore approach
PR #1905 - Task result streaming proposal

The key blocker is that JSON-RPC doesn't natively support streaming responses, so the spec needs to define how this works first.

Closing since this requires spec changes - if you're interested in this feature, I'd encourage following or contributing to the spec discussions above.

_{AI Disclaimer}

Navigation Menu

Repository navigation

can we call tools and return steaming output? #472

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

can we call tools and return steaming output? #472

Description

Activity

guanyang-nb commented on Apr 11, 2025

shihanmax commented on Apr 11, 2025

reqyou commented on Jun 25, 2025

artursanntos commented on Jun 25, 2025

lloorree commented on Aug 18, 2025

felixweinberger commented on Oct 6, 2025

leohuai commented on Dec 15, 2025

maxisbey commented on Dec 15, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions