Getting Intermediate responses from MCP Tool Calls

[Rockyyost]

Question

Does Pydantic AI have a way to receive streamed events/message during a tool call operation on an MCP Server?

Using the python lib FastMCP, looks like I can "yield" messages or use ctx.info to send string back.

What I'd like to do is send back to the user (or the LLM) any messages the tool is sending back before it gets the final response.

Additional Context

FastMCP 2.3.3
Pydantic AI 0.1.8
Python 3.10.9

[DouweM]

@Rockyyost MCP progress notifications are not currently supported by PydanticAI, as you've found. Can you please share some example code of an MCP server that uses these, so we can start thinking about how it'd make sense to surface these in PydanticAI?

[lachlanritchie1]

I am running into this issue too. I have a tool which makes a call a seperate external system and requires the user to approve something on that system. The it automatically triggers another final task.

I need to to look like this:

1. Message to user "Successfully triggered action. Please approve on the external system"
2. User approves on extrenal system
3. Message to user "Confirming apporval on the extrernal system"
4. Next action triggered.
5. Final confirmation to user that the end to end task was successful.

Right now I can only show step5 to the user. How can I show progress messages to the user (as in steps 1 and 3 before hand)?

Please note I also tried seperating out the steps above in to seperate MCP tools and I was able to easly chain them in series but the user messages won't display until step 5 no matter what I try.

Pleae help! Thank you

[DouweM]

@lachlanritchie1 Is your MCP server using progress updates? Check out #3416 where