progressToken type mismatch breaks progress callback between spring-ai and python-sdk

[myifeng]

Initial Checks

• I confirm that I'm using the latest version of MCP Python SDK
• I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue

Description

Environment

Spring AI version: 1.1.0-RC1

mcp version: 1.20.0

Language: Java (Spring Boot) + Python (async client)

Description

I'm testing the progress callback feature in the MCP protocol, but no progress notifications are received on the Python client side.
The tool executes successfully and returns "success", but the registered progress_callback function is never triggered.

Example Code

@McpTool(description = "Test progress callback.")
public String testProgress(McpSyncRequestContext context, @McpProgressToken String progressToken) {
    log.info("progressToken: {}", progressToken);

    if (progressToken != null) {
        context.progress(p -> p.progress(0.0).total(1.0).message("Starting analysis"));
    }

    if (progressToken != null) {
        context.progress(p -> p.progress(1.0).total(1.0).message("Analysis complete"));
    }

    return "success";
}

Expected behavior:

The progress messages (“Starting analysis” and “Analysis complete”) should be sent to the client.

Actual behavior:

The progressToken is not null, and both context.progress(...) calls are executed (confirmed by logs).

However, no progress messages are received on the Python side.

async def handle_progress_message(progress: float, total: float | None, message: str | None) -> None:
    """Handle progress messages"""
    print(f"\r🔍 {message}", end='', flush=True)
    print()

async def call_tool(self, name: str, args: dict) -> Any:
    loop = self._worker_task.get_loop()
    coro = self.session.call_tool(name, progress_callback=handle_progress_message, arguments=args)

    if loop is asyncio.get_running_loop():
        return await coro

    fut: concurrent.futures.Future = asyncio.run_coroutine_threadsafe(coro, loop)
    return await asyncio.wrap_future(fut)

Expected behavior:

The function handle_progress_message() should be called twice with messages from the server.

Actual behavior:

The function is never called.

The tool call completes successfully and returns the "success" string, but no progress updates are printed.

Question

Is there any known limitation or additional setup required to enable progress callbacks between Spring AI’s MCP implementation and mcp-python?

Do I need to:

Enable streaming or a specific event type in the connection?

Return progress updates differently from the Java side?

Any example or clarification on how progress events are supposed to flow between these two implementations would be very helpful.

### Python & MCP Python SDK

```Text
Spring AI version: 1.1.0-RC1

mcp version: 1.20.0

[myifeng]

@McpTool(description = "get_weather", name = "get_weather")
    public String getWeather(McpSyncRequestContext context, @McpProgressToken String progressToken) {
        // Access progress token from context
        log.info("progressToken: {}", progressToken);

        if (progressToken != null) {
            context.info("Processing data: 0");
            context.progress(p -> p.progress(0.0).total(1.0).message("Starting analysis"));
        }

        if (progressToken != null) {
            context.info("Processing data: 1");
            context.progress(p -> p.progress(1.0).total(1.0).message("Analysis complete"));
        }
        return "success";
    }

import asyncio

from mcp import ClientSession
from mcp.client.sse import sse_client
from mcp.types import LoggingMessageNotificationParams


async def logging_callback(params: LoggingMessageNotificationParams):
    print(f"\n[Server Log - {params.level.upper()}] {params.data}")
    
async def progress_callback(progress: float, total: float | None, message: str | None) -> None:
    print(f"\n[Progress {progress}/{total}]: {message}")

async def main():
  async with sse_client("http://localhost:8080/sse") as (read_stream, write_stream):
       async with ClientSession(
           read_stream, 
           write_stream, 
           logging_callback=logging_callback
        ) as session:
            await session.initialize()
            response = await session.list_tools()
            tools = response.tools
          
            # Call get_weather tool
            weather_result = await session.call_tool(
                name="get_weather", 
                arguments={"latitude": 32.0853, "longitude": 34.7818}, 
                progress_callback=progress_callback
            )
            print("\nWeather:", weather_result)

if __name__ == "__main__":
    asyncio.run(main())

Only Log, no progress

[Server Log - INFO] Processing data: 0

[Server Log - INFO] Processing data: 1

Weather: meta=None content=[TextContent(type='text', text='success', annotations=None, meta=None)] structuredContent=None isError=False

[myifeng]

Root cause: Type mismatch for progressToken. The Python client sends progressToken as a number (e.g. 2), but spring-ai converts/returns it as a string ("2"). The mcp-python SDK maps progress callbacks using numeric keys, so when the server returns a string token the SDK fails to find the registered callback and progress_callback is never invoked.

Server:

Client:

SessionMessage(message=JSONRPCMessage(root=JSONRPCNotification(method='notifications/progress', params={'progressToken': '2', 'progress': 0.0, 'total': 1.0, 'message': 'Starting analysis', '_meta': {}}, jsonrpc='2.0')), metadata=None)

[maxisbey]

Unless I'm missing something here, it seems like Spring AI shouldn't be converting this into a string and instead treat this as an opaque token?

[myifeng]

Unless I'm missing something here, it seems like Spring AI shouldn't be converting this into a string and instead treat this as an opaque token?

This seems to be the conversion problem of java-sdk, and spring ai is also dependent on java-sdk. Should we consider how to unify the difference between python-sdk and java-sdk according to the MCP protocol?

In addition, progressToken should be customized by manually setting _meta, but in python-sdk, request_id is forced here, even if _meta is manually defined.

[myifeng]

modelcontextprotocol/java-sdk#659