Add support for surfacing MCP tool progress updates back to user

[alamkanak]

Description

MCP tools support progress updates via ctx.report_progress
Reference: https://gofastmcp.com/servers/progress

However, these progress updates are not surfaced to the user when using pydantic ai agents
Please add support for progress updates in the agents

References

https://gofastmcp.com/servers/progress

[DouweM]

@alamkanak How would you like to handle these progress updates on the Pydantic AI side? With a custom progress callback, or some kind of event?

Note that the FastMCPToolset (https://ai.pydantic.dev/mcp/fastmcp-client/) takes a fastmcp.Client that already supports handling progress: https://gofastmcp.com/clients/progress

[alamkanak]

No hard opinion about how it should be implemented. But something like the existing elicitation_callback could work.
e.g.

async def handle_progress(
    context: RequestContext,
    params: ProgressParams,
):
    # handle progress updates here
    ...

server = MCPServerStdio(
    command='uvx',
    args=['mcp-run-python', 'stdio'],
    elicitation_callback=handle_elicitation,
    progress_callback=handle_progress 
)

Also, FastMCPToolset seems to take a server thats in the same codebase. But what about remote MCP servers?

[DouweM]

@alamkanak FastMCPToolset takes a whole bunch of things, including URLs: https://ai.pydantic.dev/mcp/fastmcp-client/#usage

[alamkanak]

I see. Perfect @DouweM. That works. Thanks. Appreciate it.